Awesome RGBDT Manipulation
← Back to the library

Broad coverage. Explicit boundaries.

A reviewed map of the field through 21 Oct 2026, including recent work and the foundations needed to interpret it. Source availability and experimental scope remain visible.

53benchmark records
606policy records
721unique research records

Some papers contribute to both collections, so collection counts cannot be added. 12 supporting resources are accessible under Tools & data. 54 supporting models and 9 surveys appear in relevant categories and are counted separately from policies. Models describe representation/prediction interfaces and downstream evaluation; surveys describe literature scope and taxonomy. 721 records have locally stored, credited figures; 2007 distinct primary-source URLs support the catalog.

Benchmark coverage

Research areaScopeRecords
Visuo-tactileBenchmarks explicitly evaluating touch-conditioned perception or control.22
Precision & assemblyInsertion, assembly, tolerance and contact-intensive evaluation.26
DexterousMulti-finger, in-hand and dexterous bimanual evaluation.14
General manipulationBroader manipulation suites with relevant tasks and evaluation infrastructure.13
RobustnessGeneralization, perturbation, memory and evaluation diagnostics.15
Tools & dataSupporting simulators, datasets and perception resources; not counted as closed-loop benchmarks.12

Policy coverage

Research areaScopeRecords
Visuo-tactile learningImitation, diffusion, ACT and reactive policies using touch or force feedback. Cutaneous tactile and wrist force/torque remain distinct inputs.78
Tactile VLA & WAMLanguage-conditioned, foundation and world-action methods using touch or force. Training-only supervision and deployment observations remain distinct.110
RGBD & 3DPolicies using observed depth or 3D representations. Measured and predicted geometry are distinguished in details.168
Tactile representationsTactile encoders, multimodal representations and pretraining models. Their downstream use is distinguished from an executable action policy.43
Geometry & world modelsDepth, spatial and predictive models relevant to manipulation. Measured inputs, estimated geometry and generated futures retain their separate roles.23
Visual precision & RLVisual precision policies, reinforcement learning and policy refinement.44
Agent / RSI / ICLRobot agents, feedback-driven self-improvement, in-context adaptation and related surveys. Method records identify what changes and how robot actions are executed; surveys describe their literature scope.261
Generalist foundationsSelected VLA and world-action foundations relevant to manipulation evaluation.142

Categories can overlap. They are discovery routes, not mutually exclusive claims about an architecture or sensor.

From benchmark rosters to policies

Paper baselines, current leaderboard participants and integration registries are checked separately. The mappings below retain the names used by each source. An integration indicates adapter support; it does not establish a leaderboard result. Counts are source aliases and can overlap across sources or map to the same research record.

Primary sourceSource typeNames mapped
RoboTwin 2.0 paper policy comparison ↗Paper baselines5 / 5
RoboTwin2 leaderboard: joint 50-task ↗Leaderboard snapshot19 / 20
RoboTwin2 leaderboard: single-task ↗Leaderboard snapshot5 / 5
RoboDojo paper: simulation baselines ↗Paper baselines30 / 30
RoboDojo paper: real-robot baselines ↗Paper baselines10 / 10
RoboDojo live leaderboard: simulation ↗Leaderboard snapshot60 / 60
RoboDojo live leaderboard: real robots ↗Leaderboard snapshot11 / 11
XPolicyLab integrated adapter registry ↗Integration registry68 / 69
N₀-Foundation NeoReal / NeoSim baseline rosters ↗Paper baselines9 / 9
Evo-DexBench evaluated policies and No-Touch ablation ↗Paper baselines6 / 6
ForceBench simulation comparison ↗Paper baselines9 / 10
SparkArena simulation / hardware paper baselines ↗Paper baselines11 / 13
RoboDojo official Agent / Agent+VLA / Agent+WAM track ↗Leaderboard snapshot7 / 7
RoboTwin 2.0 paper policy comparison — model mappings

Section 4.5: 5 policies on 50 tasks, Aloha-AgileX, 50 clean demonstrations/task, clean/randomized evaluation. Separate data-generation LLMs are outside this policy roster.

RoboTwin2 leaderboard: joint 50-task — model mappings

Official snapshot updated 2026-10-11; Aloha-AgileX, 2,500 clean demos, 100 trials/task in clean/randomized settings; one checkpoint jointly trained on 50 tasks.

RoboTwin2 leaderboard: single-task — model mappings

Official snapshot updated 2026-10-11; Aloha-AgileX, 2,500 clean demos, 100 trials/task in clean/randomized settings; separate checkpoint fine-tuned per task.

RoboDojo paper: simulation baselines — model mappings

Table 1 frozen paper snapshot (July 3 cutoff), 42 tasks and five capability dimensions; per-task success / continuous score. HumanExpert reference excluded from model count.

RoboDojo paper: real-robot baselines — model mappings

Real evaluation: 10 policies, 18 tasks across ARX X5, Piper, PiperX, 10 trials/task/robot setting; full paper protocol rather than later live board.

RoboDojo live leaderboard: simulation — model mappings

Official generated snapshot updated2026-10-20. Simulation60rows, including 7 Agent-track submissions; real11rows. Model labels, variant checkpoints and per-track membership are preserved, separate from frozen paper baselines.

  • EmbodiedEvo → EmbodiedEvo
  • VPP2 → Video Prediction Policy 2 (VPP2) — Final-name leaderboard asset; do not merge its score with VPP2-Preview.
  • PhysicalRSI → PhysicalRSI 1.0
  • Awomo-0.5 → Awomo-0.5 — Identifiable official project / author-contributed model release; manuscript or original architecture figure not provided. Training and release limitations are preserved in the policy record.
  • Simate-beta → Simate-beta — Identifiable official project / author-contributed model release; manuscript or original architecture figure not provided. Training and release limitations are preserved in the policy record.
  • Rex-M1 Preview → Rex-M1 Preview — Resolved from the author-contributed long-context adapter and separately released ModelScope checkpoint; distinct from the upstream Xiaomi-Robotics-1 base.
  • VPP2-Preview → Video Prediction Policy 2 (VPP2) — Preview leaderboard asset associated with VPP2; retain its own score and checkpoint scope.
  • InternW0-Δ → InternW0-Δ
  • Liber-0 Preview → Liber-0 Preview — Resolved from the author-contributed hierarchical adapter and separately released Preview/context checkpoints; distinct from Liber0-Lite.
  • Liber-0 Lite → Liber0-Lite — Identifiable official project / author-contributed model release; manuscript or original architecture figure not provided. Training and release limitations are preserved in the policy record.
  • GPT-6-Astra → RoboProbe / GPT-6 Astra Direct-EEF — RoboProbe frozen API policy under Direct-EEF action interface; model variant is part of the experimental setting.
  • DM0.5 → DM0.5 / OpenDM
  • ME-Brain-1.0 → ME-Brain-1.0
  • GalaxeaVLA (G0.5) → Galaxea G0.5 — Distinct G0.5 technical report / checkpoint generation.
  • Xiaomi-Robotics-1 → Xiaomi-Robotics-1
  • MachEmbodied-U0 → MachEmbodied-U0 (ME-U0)
  • OpenWAM-α → OpenWAM / OpenWAM-α
  • Meituan-Robotics-0 → StarVLA codebase — Evaluation-only StarVLA-family checkpoint (Qwen3.5-4B / MMDiT-Psi0); no distinct paper released in adapter. Mapping describes implementation lineage, not identical weights/scores.
  • HWM-CogWAM → CogWAM
  • Hy-Embodied-0.5-VLA → Hy-Embodied-0.5-VLA
  • KinRT → KinRT
  • SimpleMemVLA → SimpleMemVLA
  • Spatial Forcing → Spatial Forcing
  • Pi-05 → π0.5
  • InternVLA-A1.5 → InternVLA-A1.5
  • StarVLA-PI_v3 → StarVLA codebase — Named StarVLA configuration; share codebase identity while retaining distinct head/checkpoint and evaluated score.
  • VLAct → VLAct — Resolved official StarVLA team paper2608.27550 with released backbone and attached executable policies; distinct from VLAC action-critic.
  • X-VLA → X-VLA
  • MolmoAct2 → MolmoAct2
  • StarVLA-GR00T → StarVLA codebase — Named StarVLA configuration; share codebase identity while retaining distinct head/checkpoint and evaluated score.
  • StarVLA-OFT → StarVLA codebase — Named StarVLA configuration; share codebase identity while retaining distinct head/checkpoint and evaluated score.
  • X-WAM → X-WAM
  • Xiaomi-Robotics-0 → Xiaomi-Robotics-0
  • StarVLA → StarVLA codebase — Family / codebase label. α, OFT, GR00T and PI configurations have distinct recipes and scores; exact leaderboard head is not inferred where unspecified.
  • GigaWorld-Policy-0 → GigaWorld-Policy — Original GigaWorld-Policy family; distinct from GigaWorld-Policy-0.5 release.
  • GalaxeaVLA (G0) → Galaxea G0 — Original G0 family; RoboDojo paper recipe uses a G0Plus-3B checkpoint, not a claim that all G0 assets are equivalent.
  • LingBot-VLA → LingBot-VLA
  • EventVLA → EventVLA
  • AHA-WAM → AHA-WAM
  • ABot-M0 → ABot-M0
  • Fast-WAM → Fast-WAM
  • Pi-0 → π0
  • DeepSeek-Flash → RoboProbe / GPT-6 Astra Direct-EEF — RoboProbe baseline API-model variant under the same Direct-EEF interface; distinct from a learned VLA checkpoint.
  • GROOT-N1.7 → GR00T N1.7
  • InternVLA-A1 → InternVLA-A1
  • SmolVLA (Single Task) → SmolVLA — Per-task fine-tuned checkpoint setting, not an independent paper.
  • LDA-1B → LDA-1B
  • GPT-5.5 → RoboProbe / GPT-6 Astra Direct-EEF — RoboProbe baseline API-model variant evaluated with the same Direct-EEF interface; no separate robotics paper inferred.
  • ACT (Single Task) → ACT — Per-task trained checkpoint setting, not an independent paper.
  • GO-1 → GO-1 / AgiBot World
  • H-RDT → H-RDT
  • RDT → RDT-1B
  • DM0 → DM0
  • Dexora-1B → Dexora
  • A1 → A1
  • Spirit v1.5 → Spirit v1.5
  • TinyVLA → TinyVLA
  • OpenVLA-OFT → OpenVLA-OFT / OFT+
  • WorldScape-R²SI → WorldScape R²SI — Distinct R²SI simulation submission (2026-10-17 snapshot). Identity/team/results are official; architecture and improvement descriptions remain attributed launch coverage, without inheriting WorldScape Policy 2.0 artifacts.
  • HexaShell-Preview → HexaShell / Preview — Agent track; official Score53.18/SR46.85%. Frozen RGB tools and partial harness artifact release, not a measured-depth or learned-policy checkpoint.
RoboDojo live leaderboard: real robots — model mappings

Official generated snapshot updated2026-10-20. Simulation60rows, including 7 Agent-track submissions; real11rows. Model labels, variant checkpoints and per-track membership are preserved, separate from frozen paper baselines.

XPolicyLab integrated adapter registry — model mappings

69 actual integrated policy adapters at commit 5d0be1499634d3176258c8f2e31f5fe816497b0a, reviewed October 21; demo_policy excluded. Integration/serving is not evidence of evaluated RoboDojo/RoboTwin performance.

  • Mem-0 → Mem-0 — Learned reference policy introduced by RMBench; one cross-listed benchmark/policy work, with separate retrained July checkpoints.
  • StarVLA → StarVLA codebase — Family / codebase label. α, OFT, GR00T and PI configurations have distinct recipes and scores; exact leaderboard head is not inferred where unspecified.
  • A1 → A1
  • AHA-WAM → AHA-WAM
  • ABot-M0 → ABot-M0
  • Being-H05 → Being-H0.5
  • DM0 → DM0
  • Dexora-1B → Dexora
  • DreamZero → DreamZero
  • EventVLA → EventVLA
  • FastWAM → Fast-WAM
  • G0 → Galaxea G0
  • G0.5 → Galaxea G0.5
  • GO-1 → GO-1 / AgiBot World
  • GR00T-N1.7 → GR00T N1.7
  • GigaWorld-Policy → GigaWorld-Policy
  • H-RDT → H-RDT
  • Hy-Embodied-0.5-VLA → Hy-Embodied-0.5-VLA
  • InternVLA-A1 → InternVLA-A1
  • InternVLA-A1.5 → InternVLA-A1.5
  • LDA-1B → LDA-1B
  • LingBot-VA → LingBot-VA
  • LingBot-VLA → LingBot-VLA
  • Meituan-Robotics-0 → StarVLA codebase — Evaluation-only StarVLA-family checkpoint (Qwen3.5-4B / MMDiT-Psi0); no distinct paper released in adapter. Mapping describes implementation lineage, not identical weights/scores.
  • MolmoAct2 → MolmoAct2
  • OLA-SEM → UniWAM / OLA-Sem — Official RoboTwin news (2026-10-10) renames OLA-Sem to UniWAM; this is an alias, not a second paper.
  • OpenDM → DM0.5 / OpenDM
  • OpenVLA-OFT → OpenVLA-OFT / OFT+
  • OpenWAM → OpenWAM / OpenWAM-α
  • π0 → π0
  • π0.5 → π0.5
  • π0-Fast → π0-FAST / FAST
  • RDT-1B → RDT-1B
  • RISE → RISE (Compositional World Model) — OpenDriveLab self-improving policy2602.11075; distinct from the RGB-D RISE2404.12281 work.
  • SmolVLA → SmolVLA
  • Spatial Forcing → Spatial Forcing
  • Spirit v1.5 → Spirit v1.5
  • TinyVLA → TinyVLA
  • X-VLA → X-VLA
  • X-WAM → X-WAM
  • Xiaomi-Robotics-0 → Xiaomi-Robotics-0
  • Xiaomi-Robotics-1 (XR-1) → Xiaomi-Robotics-1
  • ACT → ACT
  • DP → Diffusion Policy
  • Evo-1 → Evo-1
  • InternW0_delta → InternW0-Δ
  • KinRT → KinRT
  • PatchWAM-Lite → PatchWAM / PatchWAM-Lite — Integrated Lite checkpoint/configuration; shares the PatchWAM paper identity, not necessarily the full model cost.
  • Awomo-0.5 → Awomo-0.5 — Identifiable official project / author-contributed model release; manuscript or original architecture figure not provided. Training and release limitations are preserved in the policy record.
  • MoPA → MoPA
  • PhysicalRSI → PhysicalRSI 1.0
  • Simate-beta → Simate-beta — Identifiable official project / author-contributed model release; manuscript or original architecture figure not provided. Training and release limitations are preserved in the policy record.
  • SimpleMemVLA → SimpleMemVLA
  • WorldScape Policy 2.0 → WorldScape Policy 2.0
  • GPT-6-Astra-Direct-EEF → RoboProbe / GPT-6 Astra Direct-EEF — RoboProbe frozen API policy with Direct-EEF interface and local Cartesian planner.
  • VPP2 → Video Prediction Policy 2 (VPP2) — Final-name leaderboard asset; do not merge its score with VPP2-Preview.
  • Liber_0_lite → Liber0-Lite — Identifiable official project / author-contributed model release; manuscript or original architecture figure not provided. Training and release limitations are preserved in the policy record.
  • Discrete Forcing → Discrete Forcing
  • EgoVLA → EgoVLA
  • MM-ABC → MM-ABC — Identifiable official project / author-contributed model release; manuscript or original architecture figure not provided. Training and release limitations are preserved in the policy record.
  • HWM-CogWAM → CogWAM
  • ME-Brain-1.0 → ME-Brain-1.0
  • ME-U0 → MachEmbodied-U0 (ME-U0)
  • Liber-0 Preview → Liber-0 Preview — Separate Preview checkpoint and hierarchical context controller, sharing the Lite runtime.
  • ME-Dex 1.0 → ME-Dex 1.0 — Existing ME-Dex manuscript and official source; the released RoboTwin interface zeros observed tactile, separately from tactile replay supervision.
  • Rex-M1 Preview → Rex-M1 Preview — Independent long-context derivative checkpoint and released inference interface.
  • Spark-0 → Spark-0 — Existing canonical Spark-0; standalone source is MIT, adapter and vendored components retain their own licenses.
  • ViGAR → ViGAR — Existing canonical ViGAR; adapter directly cites the same paper and author repository.
  • CloudRobo_WM — New adapter integration only; unambiguous original work and independent primary technical report are not established. No scored leaderboard result or speculative new paper is created.
N₀-Foundation NeoReal / NeoSim baseline rosters — model mappings

Nine representative families described for NeoReal; NeoSim table evaluates seven (excluding ACT and Diffusion Policy). NeoForce-conditioned variants are not independent foundation-model identities.

  • ACT → ACT — Paper baseline identity; sensor integration and evaluation setting remain specific to the N₀-Foundation experiment.
  • Diffusion Policy → Diffusion Policy — Paper baseline identity; sensor integration and evaluation setting remain specific to the N₀-Foundation experiment.
  • π0.5 → π0.5 — Paper baseline identity; sensor integration and evaluation setting remain specific to the N₀-Foundation experiment.
  • Xiaomi-Robotics-0 → Xiaomi-Robotics-0 — Paper baseline identity; sensor integration and evaluation setting remain specific to the N₀-Foundation experiment.
  • InternVLA-A1 → InternVLA-A1 — Paper baseline identity; sensor integration and evaluation setting remain specific to the N₀-Foundation experiment.
  • StarVLA-α → StarVLA-α — Paper baseline identity; sensor integration and evaluation setting remain specific to the N₀-Foundation experiment.
  • Fast-WAM → Fast-WAM — Paper baseline identity; sensor integration and evaluation setting remain specific to the N₀-Foundation experiment.
  • GigaWorld Policy → GigaWorld-Policy — Paper baseline identity; sensor integration and evaluation setting remain specific to the N₀-Foundation experiment.
  • LingBot-VA → LingBot-VA — Paper baseline identity; sensor integration and evaluation setting remain specific to the N₀-Foundation experiment.
Evo-DexBench evaluated policies and No-Touch ablation — model mappings

Five named policy configurations over15 evaluated tasks; No-Touch is an additional KineDex ablation. DP-C and DP-T share a canonical Diffusion Policy paper; the benchmark adapter is not the original KineDex sensor interface.

  • π0.5 → π0.5 — Paper baseline, separate single/bimanual checkpoint settings.
  • SmolVLA → SmolVLA — Paper baseline with benchmark interface adaptations.
  • DP-C → Diffusion Policy — Convolutional temporal U-Net variant.
  • DP-T → Diffusion Policy — Transformer variant; same canonical policy paper.
  • KineDex → KineDex — Original policy independently reviewed;123-taxel benchmark adaptation is separate.
  • KineDex No-Touch → KineDex — Removes tactile inputs, force supervision and force feedback; not an isolated input-only ablation.
ForceBench simulation comparison — model mappings

Ten evaluated simulation systems; real hardware includes eight. New force-interface variants share one canonical paper; external policy budgets and action horizons differ.

SparkArena simulation / hardware paper baselines — model mappings

Named main tables span vision-only simulation and separate hardware studies; baseline rosters are track-dependent. Pretraining variants share canonical Spark-0.

  • π0.5 → π0.5 — Named evaluated baseline; track-specific training and observations follow the Spark-0 tables, not every benchmark track.
  • GR00T-N1.7 → GR00T N1.7 — Named evaluated baseline; track-specific training and observations follow the Spark-0 tables, not every benchmark track.
  • StarVLA → StarVLA codebase — Named evaluated baseline; track-specific training and observations follow the Spark-0 tables, not every benchmark track.
  • RDT-1B → RDT-1B — Named evaluated baseline; track-specific training and observations follow the Spark-0 tables, not every benchmark track.
  • X-VLA → X-VLA — Named evaluated baseline; track-specific training and observations follow the Spark-0 tables, not every benchmark track.
  • EgoVLA → EgoVLA — Named evaluated baseline; track-specific training and observations follow the Spark-0 tables, not every benchmark track.
  • H-RDT → H-RDT — Named evaluated baseline; track-specific training and observations follow the Spark-0 tables, not every benchmark track.
  • FastWAM → Fast-WAM — Named evaluated baseline; track-specific training and observations follow the Spark-0 tables, not every benchmark track.
  • ACT → ACT — Named evaluated baseline; track-specific training and observations follow the Spark-0 tables, not every benchmark track.
  • FTP-1 → FTP-1 — Named evaluated baseline; track-specific training and observations follow the Spark-0 tables, not every benchmark track.
  • VITRA — Named human-pretraining baseline; exact primary identity and independent technical record remain unresolved.
  • UniT (human-action policy) — Human-action baseline is not silently mapped to the distinct UniT tactile representation paper; original policy identity remains unresolved.
  • Spark-0 Scratch / Visual PT / 100H / 1000H → Spark-0 — Within-paper variants and pretraining budgets, not additional canonical papers.
RoboDojo official Agent / Agent+VLA / Agent+WAM track — model mappings

October 20 JSON and official app system-type map;7 submission aliases map to 5 canonical robot systems. Three base-LLM rows share the RoboProbe Direct-EEF adapter. Agent labels do not imply measured depth, neural-weight updates or complete source release.

  • EmbodiedEvo → EmbodiedEvo — Official system type agent; Score 45.08, success 38.95%. Observation/adaptation/release mechanism is reviewed from the original robot system, not inferred from this type label.
  • PhysicalRSI → PhysicalRSI 1.0 — Official system type agent-vla; Score 36.27, success 31.38%. Observation/adaptation/release mechanism is reviewed from the original robot system, not inferred from this type label.
  • GPT-6-Astra → RoboProbe / GPT-6 Astra Direct-EEF — Official system type agent; Score 28.97, success 22.48%. Observation/adaptation/release mechanism is reviewed from the original robot system, not inferred from this type label.
  • DeepSeek-Flash → RoboProbe / GPT-6 Astra Direct-EEF — Official system type agent; Score 2.99, success 1.92%. Observation/adaptation/release mechanism is reviewed from the original robot system, not inferred from this type label.
  • GPT-5.5 → RoboProbe / GPT-6 Astra Direct-EEF — Official system type agent; Score 1.13, success 0.88%. Observation/adaptation/release mechanism is reviewed from the original robot system, not inferred from this type label.
  • WorldScape-R²SI → WorldScape R²SI — Official system type agent-wam; Score 46.58, success 39.35%. Observation/adaptation/release mechanism is reviewed from the original robot system, not inferred from this type label.
  • HexaShell-Preview → HexaShell / Preview — Official system type agent; Score 53.18, success 46.85%. Observation/adaptation/release mechanism is reviewed from the original robot system, not inferred from this type label.

Download the source-to-record mappings ↗

From tactile and geometry sources to records

Collections, surveys, recent papers and official releases help discover omissions. Each indexed record retains its own primary evidence. The counts below refer to source aliases, which can overlap and do not measure unique policies. Exclusions and candidates stay visible.

Discovery sourceScoped branchIndexed / aliases
N₀-Foundation original report and official release ↗NeoForce representation, NeoReal/NeoSim benchmark roles and current release artifacts, verified against the report and implementation.1 / 1
Awesome-Touch / Visual-Tactile ↗Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.4 / 19
Awesome-Touch / Data Collection ↗Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.24 / 37
Awesome-Touch / VLA/WM/LLM/VLM ↗Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.30 / 31
Awesome-Touch / Dexterous Manipulation ↗Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.16 / 21
Awesome-Touch / Manipulation/Grasping ↗Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.30 / 111
User seed ↗Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.1 / 1
Primary source expansion / representation cross-review ↗Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.4 / 5
Primary expansion / official author site ↗Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.2 / 2
Awesome-Touch representations / reviews and primary-source expansion ↗Representation Learning and Review source sections plus separately discovered tactile models, datasets and surveys. Adjacent resources are retained without manufacturing an action policy.38 / 67
RynnWorld-4D, geometry surveys and original method bibliographies ↗Depth, 3D geometry and geometry-grounded manipulation models; primary methods read independently of discovery lists.54 / 61
World Models for Robotic Manipulation — geometry reference subset ↗Eight named 3D/4D and geometry/action works in the survey, each reviewed through its original primary paper. This is a scoped subset, not a claim to close all 162 references.8 / 8
Tactile survey representation/control gaps ↗Named primary-source gaps found through the indicated bibliography or benchmark baseline; original methods and current artifacts were reviewed separately.8 / 8
Tactile-guided survey control gaps ↗Named primary-source gaps found through the indicated bibliography or benchmark baseline; original methods and current artifacts were reviewed separately.10 / 10
Evo-DexBench KineDex baseline — original policy ↗Named primary-source gaps found through the indicated bibliography or benchmark baseline; original methods and current artifacts were reviewed separately.1 / 1
Recent official arXiv — scoped sensing/contact additions ↗Seven official arXiv queries returned67 unindexed candidates across broad robotics/vision topics without source errors. This table retains the sensing/contact subset inspected here; it is not67 completed primary reviews or an automatic feed.3 / 8
19 October update: arXiv, Awesome-Touch and official release follow-up ↗Recent precision/contact-rich policies and manipulation-relevant predictive models, verified against original papers; official release review is separate from discovery feeds.14 / 17
October21 primary manipulation expansion and independent Agent/RGB-D review ↗Recent arXiv cs.RO/cs.CV, official GitHub project releases, Awesome-list/daily-feed discovery and independent official leaderboard review. Entries below are indexed only after their own primary method/resource sources were inspected.23 / 26
N₀-Foundation original report and official release — 1 source aliases

NeoForce representation, NeoReal/NeoSim benchmark roles and current release artifacts, verified against the report and implementation.

Indexed canonical records (1)
Awesome-Touch / Visual-Tactile — 19 source aliases

Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.

Source snapshot: 2026-10-17

Indexed canonical records (4)
Adjacent sensing, data and review resources (3)
  • StereoTac: a Novel Visuotactile Sensor that Combines Tactile Sensing with 3D Vision source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • CompdVision: Combining Near-Field 3D Visual and Tactile Sensing Using a Compact Compound-Eye Imaging System source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • TransTac: Visuo-Tactile Modality Transition via Ultraviolet-Encoded Transparent Elastomers source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
Candidates requiring separate source review (12)
  • Combining Finger Vision and Optical Tactile Sensing: Reducing and Handling Errors While Cutting Vegetables source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Robotic grasp control with high-resolution combined tactile and proximity sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Seeing Through your Skin: Recognizing Objects with a Novel Visuotactile Sensor source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • UVtac: Switchable UV Marker-Based Tactile Sensing Finger for Effective Force Estimation and Object Localization source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Finger-STS: Combined Proximity and Tactile Sensing for Robotic Manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • SpecTac: A Visual-Tactile Dual-Modality Sensor Using UV Illumination source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • TIRgel: A Visuo-Tactile Sensor With Total Internal Reflection Mechanism for External Observation and Contact Detection source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • VisTac Toward a Unified Multimodal Sensing Finger for Robotic Manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Vision-based Proximity and Tactile Sensing for Robot Arms: Design, Perception, and Control source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Biomimetic multimodal tactile sensing enables human-like robotic perception source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Design and application of multimodal visual-tactile sensor for object information perception source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • FruitTouch: A Perceptive Gripper for Gentle and Scalable Fruit Harvesting source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
Awesome-Touch / Data Collection — 37 source aliases

Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.

Source snapshot: 2026-10-17

Indexed canonical records (24)
Adjacent sensing, data and review resources (5)
  • TacCap: A Wearable FBG-Based Tactile Sensor for Seamless Human-to-Robot Skill Transfer source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • TouchAnything: A Dataset and Framework for Bimanual Tactile Estimation from Egocentric Video source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • ART-Glove: Articulated Tactile Glove for Contact-Grounded Dexterous Interaction Capture source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • VICON: Visual-Inertial-Contact based Hand-Object Tracking for Manipulation Datasets source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • TouchScale: 500 Hours of Human Vision and Touch for Visual-Tactile Learning source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
Candidates requiring separate source review (8)
  • ManiWAV: Learning Robot Manipulation from In-the-Wild Audio-Visual Data source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Bunny-VisionPro: Real-Time Bimanual Dexterous Teleoperation for Imitation Learning source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • DOGlove: Dexterous Manipulation with a Low-Cost Open-Source Haptic Force Feedback Glove source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • TactileAloha: Learning Bimanual Manipulation With Tactile Sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • exUMI: Extensible Robot Teaching System with Action-aware Task-agnostic Tactile Representation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • TacUMI: A Multi-Modal Universal Manipulation Interface for Contact-Rich Tasks source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • ViHaTeleop: A Low-Cost, Lightweight Visual-Haptic Teleoperation System for Dexterous Manipulation Learning source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • DITTO-X: Forward and Reverse Teleoperation for Dexterous Manipulation and Human Intervention source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
Awesome-Touch / VLA/WM/LLM/VLM — 31 source aliases

Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.

Source snapshot: 2026-10-17

Indexed canonical records (30)
Candidates requiring separate source review (1)
  • HapTile: A Haptic-Informed Vision-Tactile-Language-Action Dataset for Contact-Rich Imitation Learning source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
Awesome-Touch / Dexterous Manipulation — 21 source aliases

Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.

Source snapshot: 2026-10-17

Indexed canonical records (16)
Adjacent sensing, data and review resources (1)
  • Temporal Visuo-Tactile Learning for Dexterous Grasp Stability source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
Candidates requiring separate source review (4)
  • All the Feels: A dexterous hand with large-area tactile sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Sampling-based Exploration for Reinment Learning of Dexterous Manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Tactile SoftHand-A: 3D-Printed, Tactile, Highly-underactuated, Anthropomorphic Robot Hand with an Antagonistic Tendon Mechanism source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • FlashDexRetarget: Accelerating Dexterous Manipulation Data Generation through Multi-Motion Retargeting source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
Awesome-Touch / Manipulation/Grasping — 111 source aliases

Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.

Source snapshot: 2026-10-17

Indexed canonical records (30)
Adjacent sensing, data and review resources (7)
  • Learning to Detect Slip through Tactile Measures of the Contact Force Field and its Entropy source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • Integrated Object Deformation and Contact Patch Estimation from Visuo-Tactile Feedback source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • Grasp Stability Assessment Through Attention-Guided Cross-Modality Fusion and Transfer Learning source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • Embedded Object Detection and Mapping in Soft Materials Using Optical Tactile Sensing source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • AcTExplore: Active Tactile Exploration on Unknown Objects source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • Neural feels with neural fields: Visuo-tactile perception for in-hand manipulation source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
  • SlipSense: Multimodal Tactile Learning for Low-Latency and Generalized Slip Detection source ↗ — Primary paper describes sensing, representation, estimation or a data resource; no independently audited executable manipulation policy is asserted in this control batch.
Candidates requiring separate source review (74)
  • Localization and manipulation of small parts using GelSight tactile sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Reliable object handover through tactile force sensing and effort control in the Shadow Robot hand source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Reliable robotic handovers through tactile sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Tactile-Based Insertion for Dense Box-Packing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Maintaining Grasps within Slipping Bounds by Monitoring Incipient Slip source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Manipulation by feel: Touch-based control with deep predictive models source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Building a Library of Tactile Skills Based on FingerVision source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Self-Attention Based Visual-Tactile Fusion Learning for Predicting Grasp Outcomes source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Deep Reinforcement Learning for Tactile Robotics: Learning to Type on a Braille Keyboard source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Swingbot: Learning physical features from in-hand tac?tile exploration for dynamic swing-up manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Tactile Dexterity: Manipulation Primitives with Tactile Feedback source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Cable manipulation with a tactile-reactive gripper source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • A Robust Controller for Stable 3D Pinching Using Tactile Sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Tactile-RL for Insertion: Generalization to Objects of Unknown Geometry source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Improving Grasp Stability with Rotation Measurement from Tactile Sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Dynamic Modeling of Hand-Object Interactions via Tactile Sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • SEED: Series Elastic End Effectors in 6D for Visuotactile Tool Use source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Learning Generalizable Vision-Tactile Robotic Grasping Strategy for Deformable Objects via Transformer source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Meta-Residual Policy Learning: Zero-Trial Robot Skill Adaptation via Knowledge Fusion source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Active Visuo-Tactile Interactive Robotic Perception for Accurate Object Pose Estimation in Dense Clutter source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Autonomous Learning of Page Flipping Movements via Tactile Feedback source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Where Shall I Touch? Vision-Guided Tactile Poking for Transparent Object Grasping source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Efficient Tactile Simulation with Differentiability for Robotic Manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Tactile Pose Estimation and Policy Learning for Unknown Object Manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Grasp Stability Prediction with Sim-to-Real Transfer from Tactile Sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Safely Learning Visuo-Tactile Feedback Policies in Real For Industrial Insertion source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Manipulation via Membranes: High-Resolution and Highly Deformable Tactile Sensing and Control source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Learning Self-Supervised Representations from Vision and Touch for Active Sliding Perception of Deformable Surfaces source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Active Extrinsic Contact Sensing: Application to General Peg-in-Hole Insertion source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Task-Driven In-Hand Manipulation of Unknown Objects with Tactile Sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Visual-tactile Fusion for Transparent Object Grasping in Complex Backgrounds source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Integrating High-Resolution Tactile Sensing into Grasp Stability Prediction source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Heuristic grasping of convex objects using 3D imaging and tactile sensing in uncalibrated grasping scenarios source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Skill generalization of tubular object manipulation with tactile sensing and Sim2Real learning source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Learning height for top-down grasps with the DIGIT sensor source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Deep Functional Predictive Control for Strawberry Cluster Manipulation using Tactile Prediction source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Cable Routing and Assembly using Tactile-driven Motion Primitives source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Tactile-Filter: Interactive Tactile Perception for Part Mating source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Visuo-Tactile Sensor Enabled Pneumatic Device Towards Compliant Oropharyngeal Swab Sampling source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Learning-Based Slip Detection for Dexterous Manipulation Using GelStereo Sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Enabling Robot Manipulation of Soft and Rigid Objects with Vision-based Tactile Sensors source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • TacMMs: Tactile Mobile Manipulators for Warehouse Automation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Visual-Tactile Learning of Garment Unfolding for Robot-Assisted Dressing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • TactileVAD: Geometric Aliasing-Aware Dynamics for High-Resolution Tactile Control source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Precise Robotic Needle-Threading with Tactile Perception and Reinforcement Learning source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Inter-finger Small Object Manipulation with DenseTact Optical Tactile Sensor source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • VisTac Towards a Unified Multi-Modal Sensing Finger for Robotic Manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Laboratory Automation: Precision Insertion with Adaptive Fingers utilizing Contact through Sliding with Tactile-based Pose Estimation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Tactile Active Inference Reinforcement Learning for Efficient Robotic Manipulation Skill Acquisition source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Evetac: An Event-based Optical Tactile Sensor for Robotic Manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Pose and shear-based tactile servoing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • GelStereo BioTip: Self-Calibrating Bionic Fingertip Visuotactile Sensor for Robotic Manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Perceiving Extrinsic Contacts from Touch Improves Learning Insertion Policies source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • TEXterity: Tactile Extrinsic deXterity source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Stable Object Placing using Curl and Diff Features of Vision-based Tactile Sensors source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • DexiTac: Soft Dexterous Tactile Gripping source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Generalize by Touching: Tactile Ensemble Skill Transfer for Robotic Furniture Assembly source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • SimPLE, a visuotactile method learned in simulation to precisely pick, localize, regrasp, and place objects source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • RoTipBot: Robotic Handling of Thin and Flexible Objects using Rotatable Tactile Sensors source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Jointly Understand Visual and Tactile Signals source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • In-Hand Singulation and Scooping Manipulation with a 5 DOF Tactile Gripper source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • T-TD3: A Reinforcement Learning Framework for Stable Grasping of Deformable Objects Using Tactile Prior source ↗ — Discovery-list identity conflict: this arXivID is In-Hand Singulation and Scooping, not the separately linked T-TD3 paper. Do not create a duplicate or misattribute claims.
  • Learning Task Planning from Multi-Modal Demonstration for Multi-Stage Contact-Rich Manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Bimanual In-hand Manipulation using Dual Limit Surfaces source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Extremum Seeking Controlled Wiggling for Tactile Insertion source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • T-TD3: A Reinforcement Learning Framework for Stable Grasping of Deformable Objects Using Tactile Prior source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Analysing the Interplay of Vision and Touch for Dexterous Insertion Tasks source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • TARS: Tactile Affordance in Robot Synesthesia for Dexterous Manipulation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Image-Based Tactile Deformation Simulation and Pose Estimation for Robot Skill Learning source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Multimodal and Force-Matched Imitation Learning With a See-Through Visuotactile Sensor source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • DIH-Tele: Dexterous In-Hand Teleoperation Framework for Learning Multiobjects Manipulation With Tactile Sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • TactileAloha: Learning Bimanual Manipulation with Tactile Sensing source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • Reactive In-Air Clothing Manipulation with Confidence-Aware Dense Correspondence and Visuotactile Affordance source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
  • exUMI: Extensible Robot Teaching System with Action-aware Task-agnostic Tactile Representation source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
User seed — 1 source aliases

Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.

Source snapshot: 2026-10-17

Indexed canonical records (1)
  • Spark-0 source ↗ — Primary paper and full implementation/evaluation sections reviewed; included in this expansion.
Primary source expansion / representation cross-review — 5 source aliases

Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.

Source snapshot: 2026-10-17

Indexed canonical records (4)
Candidates requiring separate source review (1)
  • Just Add Force source ↗ — Potentially relevant control or supporting work; detailed original-source and release audit remains incomplete, so not counted as a reviewed model/policy.
Primary expansion / official author site — 2 source aliases

Named source aliases, including reviewed canonical matches, adjacent sensing/data resources and explicit candidates awaiting primary-source review. Section placement alone does not establish a deployable controller.

Source snapshot: 2026-10-17

Indexed canonical records (2)
  • SCFields source ↗ — Primary paper and full implementation/evaluation sections reviewed; included in this expansion.
  • FACT source ↗ — Primary paper and full implementation/evaluation sections reviewed; included in this expansion.
Awesome-Touch representations / reviews and primary-source expansion — 67 source aliases

Representation Learning and Review source sections plus separately discovered tactile models, datasets and surveys. Adjacent resources are retained without manufacturing an action policy.

Source snapshot: 2026-10-17

Indexed canonical records (38)
Adjacent sensing, data and review resources (26)
  • Marker or Markerless? Mode-Switchable Optical Tactile Sensing for Diverse Robot Tasks source ↗ — Mode-switchable optical sensor hardware/method paper; do not manufacture an executable policy or reusable foundation model from the list placement.
  • Artificial Sense of Slip—A Review source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Tactile sensing in dexterous robot hands — Review source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Tactile Image Sensors Employing Camera: A Review source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Tactile Sensors for Advanced Intelligent Systems source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Visuotactile Sensors With Emphasis on GelSight Sensor: A Review source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • On the Design and Development of Vision-based Tactile Sensors source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • A survey of visuotactile sensing technologies for robotic manipulation source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Hardware Technology of Vision-Based Tactile Sensor: A Review source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Marker Displacement MethodUsedinVision-Based Tactile Sensors—From 2D to 3D: A Review source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • When Vision Meets Touch: A Contemporary Review for Visuotactile Sensors from the Signal Processing Perspective source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Artificial Skin Based on Visuo-Tactile Sensing for 3D Shape Reconstruction: Material, Method, and Evaluation source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Tactile Data Generation and Applications Based on Visuo-tactile Sensors: A Review source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Tactile Robotics: An Outlook source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Classification of Vision-Based Tactile Sensors: A Review source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • A Survey of Vision-Based Tactile Sensors: Hardware, Algorithm, Application and Future Direction source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Co-evolution of Vision-Based Tactile Sensors: From structure-dominated design to task-based system integration—A Review source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • Vision-Based Tactile Intelligence for Robotics: Sensing, Learning, and Embodied Manipulation source ↗ — Hardware/signal-processing review retained as adjacent discovery; not a reviewed manipulation model entry.
  • TouchScale source ↗ — 500-hour human dataset; head RGB-D, wrist RGB and bimanual tactile streams. Public code/data links independently found. Retained as supporting dataset candidate; no new model record invented from dataset scale.
  • TouchAnything / EgoTouch source ↗ — EgoTouch captures measured dense pressure; TouchAnything predicts pressure from RGB at inference. 208tasks and1,891episodes are dataset scope, not robot-control benchmark totals. Public code and data links independently found.
  • VICON source ↗ — Human hand/object tracking and sparse normal-force capture. Evaluation uses40mocapsessions/fiveobjects; expandedten-categorydataset promised separately. Retained as acquisition/tracking candidate, not tactile policy.
  • ART-Glove source ↗ — Articulated capture hardware:22DoF and2,048taxels at120Hz. Direct human capture and contactgeometry sensing do not establish robotpolicytraining. Retained as supporting acquisition candidate.
  • Vision-Based Tactile Intelligence for Robotics source ↗ — Additional survey discovery retained for a subsequent original-source citation audit; no claim its full bibliography is covered.
  • Tactile Robotics: An Outlook source ↗ — Additional survey/outlook discovery retained for a subsequent original-source citation audit.
  • When Vision Meets Touch source ↗ — Additional signal-processing survey discovery retained for a subsequent primary-source review.
  • RLHND source ↗ — Monocular-video pose/contact/force prediction; predicted tactile information is not a measured touch input. Supporting model discovery candidate.
Candidates requiring separate source review (3)
  • Touch and Go: Learning from Human-Collected Vision and Touch source ↗ — Representation-section discovery candidate requiring distinct original-source review before inclusion.
  • Reducing Cross-Sensor Domain Gaps in Tactile Sensing via Few-Sample-Driven Style-to-Content Unsupervised Domain Adaptation source ↗ — Representation-section discovery candidate requiring distinct original-source review before inclusion.
  • A Minimal Optical-Flow Representation for Vision-Based Tactile Rotation Classification in Robotic Manipulation Across Gravity Domains source ↗ — Representation-section discovery candidate requiring distinct original-source review before inclusion.
RynnWorld-4D, geometry surveys and original method bibliographies — 61 source aliases

Depth, 3D geometry and geometry-grounded manipulation models; primary methods read independently of discovery lists.

Source snapshot: 2026-10-17

Indexed canonical records (54)
  • RynnWorld-4D / RynnWorld-4D-Policy source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • GeoAlign source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • GeoBridge-VLA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • 3D Foresight source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • 3DFlowAction source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • Canonical Policy source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • EquiForm source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • GEM3D source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • GEM-4D source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • EquivDP3 source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • PEAfowl source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • TraceGen source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • μ₀ / Trace Expert + Action Expert source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • Dex4D source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • MotionVLA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • GAP — Action-Geometry Prediction source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • TAX-DPD source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • F3RM source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • D³Fields source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • Neural Descriptor Fields (NDF) source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • Imagination Policy source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • MATCH Policy source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • Spherical Diffusion Policy (SDP) source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • DepthWorld source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • ForeAct3D source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • PointWorld + MPC source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished. Closed the geometry-model bibliography reference subset from the Wiley survey, using the original method paper rather than the survey as evidence.
  • TesserAct source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished. Closed the geometry-model bibliography reference subset from the Wiley survey, using the original method paper rather than the survey as evidence.
  • WristWorld source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished. Closed the geometry-model bibliography reference subset from the Wiley survey, using the original method paper rather than the survey as evidence.
  • Kinema4D source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished. Closed the geometry-model bibliography reference subset from the Wiley survey, using the original method paper rather than the survey as evidence.
  • Gaussian World Model (GWM) source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished. Closed the geometry-model bibliography reference subset from the Wiley survey, using the original method paper rather than the survey as evidence.
  • 3D-VLA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished. Closed the geometry-model bibliography reference subset from the Wiley survey, using the original method paper rather than the survey as evidence.
  • OG-VLA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished. Closed the geometry-model bibliography reference subset from the Wiley survey, using the original method paper rather than the survey as evidence.
  • 3D-CAVLA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished. Closed the geometry-model bibliography reference subset from the Wiley survey, using the original method paper rather than the survey as evidence.
  • LingBot-Depth source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • LingBot-Vision / LingBot-Depth 2.0 source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • RynnBrain / RynnBrain-VLA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • RynnBrain 1.1 / RynnBrain-VLA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • RynnVLA-001 source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • RynnVLA-002 source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • Bimanual RMA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • StereoPatch source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • Robo3R source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • DepthVLA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • 4D-VLA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • SwiftVLA source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • G3Flow source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • ManiGaussian source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • RoboUniView source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • DexPoint source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • EquivAct source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • Semantic-Geometric Representation (SGR) source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • SGRv2 source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • GemBench / 3D-LOTUS++ source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
  • TAX-Pose source ↗ — Original method and geometry provenance reviewed, with measured/inferred/generated geometry distinguished.
Outside this atlas's scope (4)
  • DriveDreamer-Policy source ↗ — Geometry-aware driving model; no robotic manipulation experiments.
  • VLGA source ↗ — Autonomous-driving geometry/action model, outside manipulation scope.
  • Dense4D source ↗ — Human body/clothing correspondence; no inspected manipulation policy evaluation.
  • CAP source ↗ — Humanoid locomotion under depth corruption, without manipulation experiment.
Candidates requiring separate source review (2)
  • Depth Camera or RGB-to-Depth? source ↗ — Directly relevant sensor-vs-estimator policy study located; full primary manuscript and complete release evidence not yet independently inspected.
  • DeSiRe source ↗ — University author repository verifies peer-reviewed depth-conditioned dexterous policy and date. Full text unavailable there; abstract does not resolve depth provenance, training cost or releases.
Unresolved names or unavailable sources (1)
  • RynnVLA-003 source ↗ — Exact public paper/repository identity not independently located. No speculative canonical entry created.
World Models for Robotic Manipulation — geometry reference subset — 8 source aliases

Eight named 3D/4D and geometry/action works in the survey, each reviewed through its original primary paper. This is a scoped subset, not a claim to close all 162 references.

Source snapshot: 2026-10-17

Indexed canonical records (8)
  • PointWorld + MPC source ↗ — Original model/policy paper independently reviewed; geometry provenance and downstream controller role are recorded separately.
  • TesserAct source ↗ — Original model/policy paper independently reviewed; geometry provenance and downstream controller role are recorded separately.
  • WristWorld source ↗ — Original model/policy paper independently reviewed; geometry provenance and downstream controller role are recorded separately.
  • Kinema4D source ↗ — Original model/policy paper independently reviewed; geometry provenance and downstream controller role are recorded separately.
  • Gaussian World Model (GWM) source ↗ — Original model/policy paper independently reviewed; geometry provenance and downstream controller role are recorded separately.
  • 3D-VLA source ↗ — Original model/policy paper independently reviewed; geometry provenance and downstream controller role are recorded separately.
  • OG-VLA source ↗ — Original model/policy paper independently reviewed; geometry provenance and downstream controller role are recorded separately.
  • 3D-CAVLA source ↗ — Original model/policy paper independently reviewed; geometry provenance and downstream controller role are recorded separately.
Tactile survey representation/control gaps — 8 source aliases

Named primary-source gaps found through the indicated bibliography or benchmark baseline; original methods and current artifacts were reviewed separately.

Source snapshot: 2026-10-17

Indexed canonical records (8)
  • CLTP / Tac3D-LLM source ↗ — Primary role follows the reusable central contribution; downstream control probes are retained without automatically changing representation models to policies.
  • Tac-DINO / VTPA source ↗ — Primary role follows the reusable central contribution; downstream control probes are retained without automatically changing representation models to policies.
  • TactX source ↗ — Primary role follows the reusable central contribution; downstream control probes are retained without automatically changing representation models to policies.
  • UniForce source ↗ — Primary role follows the reusable central contribution; downstream control probes are retained without automatically changing representation models to policies.
  • RA-Touch source ↗ — Primary role follows the reusable central contribution; downstream control probes are retained without automatically changing representation models to policies.
  • SToLa source ↗ — Primary role follows the reusable central contribution; downstream control probes are retained without automatically changing representation models to policies.
  • ViTaPEs source ↗ — Primary role follows the reusable central contribution; downstream control probes are retained without automatically changing representation models to policies.
  • Making Sense of Vision and Touch source ↗ — Primary role follows the reusable central contribution; downstream control probes are retained without automatically changing representation models to policies.
Tactile-guided survey control gaps — 10 source aliases

Named primary-source gaps found through the indicated bibliography or benchmark baseline; original methods and current artifacts were reviewed separately.

Source snapshot: 2026-10-17

Indexed canonical records (10)
  • TLA source ↗ — Original full text, deployment interface and available official artifact status reviewed; executable policy/controller.
  • VTLA source ↗ — Original full text, deployment interface and available official artifact status reviewed; executable policy/controller.
  • TacVLA source ↗ — Original full text, deployment interface and available official artifact status reviewed; executable policy/controller.
  • ReTouch source ↗ — Original full text, deployment interface and available official artifact status reviewed; executable policy/controller.
  • FoAR source ↗ — Original full text, deployment interface and available official artifact status reviewed; executable policy/controller.
  • ForceMimic / HybridIL source ↗ — Original full text, deployment interface and available official artifact status reviewed; executable policy/controller.
  • ImplicitRDP source ↗ — Original full text, deployment interface and available official artifact status reviewed; executable policy/controller.
  • AdapTac source ↗ — Original full text, deployment interface and available official artifact status reviewed; executable policy/controller.
  • ViTacWorld source ↗ — Original full text, deployment interface and available official artifact status reviewed; predictive model rather than executable policy.
  • TacMan-Turbo source ↗ — Original full text, deployment interface and available official artifact status reviewed; executable policy/controller.
Evo-DexBench KineDex baseline — original policy — 1 source aliases

Named primary-source gaps found through the indicated bibliography or benchmark baseline; original methods and current artifacts were reviewed separately.

Source snapshot: 2026-10-17

Indexed canonical records (1)
  • KineDex source ↗ — Original paper and official implementation verified. Original tactile interface retained independently of benchmark adapter modifications.
Recent official arXiv — scoped sensing/contact additions — 8 source aliases

Seven official arXiv queries returned67 unindexed candidates across broad robotics/vision topics without source errors. This table retains the sensing/contact subset inspected here; it is not67 completed primary reviews or an automatic feed.

Source snapshot: 2026-10-17; submittedDate20261014 onward

Indexed canonical records (3)
  • ForceBench / force-integration policies source ↗ — Found in the official arXiv Oct14–17 query, then reviewed in the complete original method/evaluation/appendix; release and figure checks completed.
  • Evo-DexBench source ↗ — Found in the official arXiv Oct14–17 query, then reviewed in the complete original method/evaluation/appendix; release and figure checks completed.
  • MAIA / VAF-WM source ↗ — Found in the official arXiv Oct14–17 query, then reviewed in the complete original method/evaluation/appendix; release and figure checks completed.
Adjacent sensing, data and review resources (1)
  • CQDC-Warp: GPU-Accelerated Differentiable Quasi-Dynamic Simulation for Contact-Rich Manipulation source ↗ — Differentiable contact simulator discovery; retained as an adjacent tool requiring separate primary review.
Candidates requiring separate source review (4)
  • TacClaw-Infinity: A Large-Scale Vision-Tactile-Language-Action Dataset for Dexterous Manipulation source ↗ — VTLA data-collection/corpus discovery; full primary-source model and release review remains pending.
  • FOMI: A Force-Feedback Manipulation Interface for Teaching Contact-Rich Manipulation source ↗ — Force-feedback teaching interface discovery; its full control contribution and releases still require separate source review.
  • TGrasp6D: Task-Informed Vision-Language Priors and Pose Refinement enable Zero-Shot Task-Oriented 6DoF Grasping source ↗ — Geometry-guided grasping candidate; full policy interface, depth provenance and release review remains pending.
  • Manifold Generative Annealing for Contact Constraint-aware Impedance Control with Reinforcement Learning Prior source ↗ — Contact/impedance control candidate; original implementation and full policy-resource review remain pending.
19 October update: arXiv, Awesome-Touch and official release follow-up — 17 source aliases

Recent precision/contact-rich policies and manipulation-relevant predictive models, verified against original papers; official release review is separate from discovery feeds.

Indexed canonical records (14)
  • Blind Precise Assembly source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • No Eyes, No Problem: Blind Baoding Manipulation source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • ORD-WM source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • Ego2Contact source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • DefFlow source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • DSP: Dexterity from Supervised Play source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • ForceWAM source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • Contact Trajectory Prompting (CTP) source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • ForeTac source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • DexNinja source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • FOMO-Dex source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • DEFT-VLA source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • Residual Compliant Policy source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
  • TacZero source ↗ — Original v1 date, full method and deployed observations verified; no release is inferred from a paper or project page.
Background references (2)
  • Stereopatch project-page source source ↗ — Repository is the Astro project page; it is not a manipulation-policy implementation release.
  • TORL-VLA linked RLT reproduction source ↗ — Linked repository describes an RLT reproduction rather than a verified TORL-VLA implementation; do not upgrade official release status from that link.
Unresolved names or unavailable sources (1)
  • DT-TacVLA source ↗ — Anonymous project presentation has no verified canonical paper/date or identifiable author provenance; held outside reviewed records.
October21 primary manipulation expansion and independent Agent/RGB-D review — 26 source aliases

Recent arXiv cs.RO/cs.CV, official GitHub project releases, Awesome-list/daily-feed discovery and independent official leaderboard review. Entries below are indexed only after their own primary method/resource sources were inspected.

Indexed canonical records (23)
  • GTA-2 source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • GraspTwin source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • TacTalent source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • HexaShell / Preview source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • CLARO source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • LAW source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • ContactKernels source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • ContactPolicy source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • MoDex source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • FR-VLA source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • Agentic RSR source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • RoboReact source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • Tactile rotation — hand morphology source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • Gen2Dex source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • ACTS / React source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • T(R,O) Grasp++ source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • NISE source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • TGrasp6D source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • Ropedia W0 source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • HandsOn source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • BiDex-100 source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • ContactBench 1.0 source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
  • MoCaP source ↗ — Original full paper and observation/control/resource scope reviewed; v1 date retained. Code, weights and data are independently scoped.
Outside this atlas's scope (2)
  • CognitiveReality source ↗ — RGB-D mapping/VR assistant with navigation/re-observation evaluation; no manipulation-control evidence established for this catalog.
  • Policy Search in Code Space for Multi-Agent Pathfinding source ↗ — Warehouse pathfinding/coordination rather than manipulation.
Adjacent sensing, data and review resources (1)
  • PhantomTac source ↗ — Tactile-system attack study; not a new manipulation policy/benchmark and not expanded into attack implementation details.

Download tactile and geometry source mappings ↗

From surveys to original methods

Surveys supply taxonomy and citation trails. Their references are reconciled below; technical claims in each indexed method still cite its own paper or official release. Background, adjacent resources and candidates remain visible without being counted as verified policies.

In-Context Learning for Robots: Methods and Applications

All 412 numbered references are reconciled. The author assigns 260 entries to four method families. This atlas follows manipulation methods; navigation, generic learning foundations and data-only resources retain explicit dispositions. A citation match is discovery provenance, not quantitative evidence for a method. Snapshot: Supplement S1, 2026-09-29-primary-source-refresh.

Author reference inventory ↗ · Author method corpus ↗

Indexed in this atlas · 239
  • One-Shot Imitation Learning · Reference 1 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Prompt-DT · Reference 5 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Meta-Imitation Learning (MIL) · Reference 7 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • DAML · Reference 8 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • In-Context Policy Improvement (ICPI) · Reference 9 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RoboTTT · Reference 10 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • WAM-TTT · Reference 11 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • MOSAIC · Reference 12 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • In-Context Robot Transformer (ICRT) · Reference 13 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Vid2Robot · Reference 14 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Keypoint Action Tokens (KAT) · Reference 15 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Instant Policy · Reference 16 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RICL · Reference 17 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • ICI-VLA · Reference 18 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • OSVI-WM · Reference 21 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • Demo-JEPA · Reference 22 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • HOST · Reference 23 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • Zero-WAM · Reference 24 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • WorldScape Policy 2.0 · Reference 25 · world · Cited source ↗

    Cited as: WorldScape Policy 2.0: Empowering Steerable World Action Modeling with Reasoning-Augmented Memory

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Neural Task Programming (NTP) · Reference 26 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Code as Policies (CaP) · Reference 27 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • ProgPrompt · Reference 28 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • XSkill · Reference 29 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • UniSkill (Human Video Imitation) · Reference 30 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Show-Harness · Reference 31 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • FlowControl · Reference 32 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • DOME · Reference 33 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • DINOBot · Reference 34 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • DITTO · Reference 35 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • MT3 · Reference 36 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • ODIL · Reference 37 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • SemAnCorr · Reference 38 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • In-Context World Modeling (ICWM) · Reference 41 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • MemoryVLA · Reference 42 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • MemER · Reference 43 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • MEM / π0.6-MEM · Reference 44 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • SimpleMemVLA · Reference 45 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • DROC · Reference 46 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • ASPIRE · Reference 47 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • ARCHITECT · Reference 48 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • SHAPER · Reference 49 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Lifelong Robot Library Learning (LRLL) · Reference 50 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • LYRA · Reference 51 · agent · Cited source ↗

    Cited as: Growing with Your Embodied Agent: A Human-in-the-Loop Lifelong Code Generation Framework for Long-Horizon Manipulation Skills

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RATs / Playful Agentic Robot Learning · Reference 52 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Teach-and-Grow Learning (TGL) · Reference 53 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Behavior Prompting Policy (BPP) · Reference 54 · policy · Cited source ↗

    Cited as: Behavior Prompting Policy: Demonstrations as Prompts for Manipulation

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • ContextFlow · Reference 55 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Open X-Embodiment / RT-X · Reference 57 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • GO-1 / AgiBot World · Reference 60 · data · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Galaxea G0 · Reference 61 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • AutoRT · Reference 76 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • SynthICL · Reference 77 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RHyME · Reference 80 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • REFLECT · Reference 84 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • KnowNo · Reference 86 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RACER · Reference 87 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Zeva · Reference 88 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • ENPIRE · Reference 89 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Zetta · Reference 90 · agent · Cited source ↗

    Cited as: Zetta zeta: An Efficient Closed-Loop Embodied Harness for Self-Evolving Physical Intelligence

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Instant-Fold · Reference 91 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • BiCICLe · Reference 92 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • ManiLong-Shot · Reference 93 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • GPT-Policy · Reference 94 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • GEN-1.5 · Reference 98 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Skild S1 · Reference 99 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • VIMA · Reference 100 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Gated Memory Policy (GMP) · Reference 105 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • LIBERO-PRO · Reference 106 · evaluation · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RoboCat · Reference 114 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Eureka · Reference 115 · training · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • DrEureka · Reference 116 · training · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Language Model Predictive Control (LMPC) · Reference 117 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • OpenVLA · Reference 118 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • π0 · Reference 119 · policy · Cited source ↗

    Cited as: π_0: A Vision-Language-Action Flow Model for General Robot Control

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • GR00T N1 · Reference 120 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • CrossFormer · Reference 132 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Kinematic Intelligence · Reference 133 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • CorrectVLA · Reference 142 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • UP-OSI · Reference 146 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Watch, Try, Learn (WTL) · Reference 147 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • BC-Z · Reference 148 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • VINN · Reference 149 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • ReCAP · Reference 152 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • RoboPrompt · Reference 153 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • REGENT · Reference 154 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • DemoDiffusion · Reference 155 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • IMOP · Reference 156 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • Task-Embedded Control Networks (TecNets) · Reference 157 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Transformers for One-Shot Visual Imitation (T-OSIL) · Reference 158 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • ViVLA · Reference 159 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • MINT · Reference 160 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • MatchingPolicy · Reference 161 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • In-Context VLA / VLA-Talker · Reference 162 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • StellaVLA · Reference 163 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Attributed Waypoints and Demonstration Augmentation (AWDA) · Reference 164 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • ORION · Reference 165 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • CaP-X / CaP-Agent0 · Reference 166 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • HRT1 · Reference 168 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • RoboTAP · Reference 169 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • Trajectory Transfer via Unseen Pose Estimation · Reference 170 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • Interaction Warping · Reference 171 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • FUNCTO · Reference 172 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • MAGIC · Reference 173 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • GIFT · Reference 174 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • Sparse Meets Dense · Reference 175 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • VLAff · Reference 176 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • AF-OSIL-MS · Reference 177 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • Universal Planning Networks · Reference 179 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • ICLR: Visual Reasoning Imitation · Reference 181 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • SeeTraceAct · Reference 182 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • PonderPounce · Reference 183 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • VoxPoser · Reference 184 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • SayCan · Reference 185 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Inner Monologue · Reference 186 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • PhysCaP · Reference 187 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • KALM · Reference 189 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Point Policy · Reference 190 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Elastic Motion Policy · Reference 191 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • SpatialVLA · Reference 193 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RTCF · Reference 195 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • ReMem-VLA · Reference 198 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • μVLA · Reference 199 · policy · Cited source ↗

    Cited as: μVLA: On Recurrent Memory for Partially Observable Manipulation in VLA Models

    Verified manipulation policy in the original primary source.

  • EventVLA · Reference 200 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Robust Instant Policy (RIP) · Reference 201 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • AMAGO · Reference 202 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • HALO · Reference 204 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • TemporalFlow-VLA · Reference 205 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Achievement-Grounded Memory (AGM) · Reference 206 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Harness VLA · Reference 207 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Q-Planning · Reference 208 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • MemoryVLA++ · Reference 209 · world · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Steerable Policies · Reference 210 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Guava-4B · Reference 211 · agent · Cited source ↗

    Cited as: Guava: An Effective and Universal Harness for Embodied Manipulation

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Pigey / Physical Agency · Reference 212 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • MALMM · Reference 213 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • PRoC3S · Reference 214 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • In-Context Abstraction Learning (ICAL) · Reference 215 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • EmbodiSkill · Reference 216 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Uni-Skill (SkillFolder) · Reference 217 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Act-Observe-Rewrite (AOR) · Reference 218 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Hierarchical VLA Design Study · Reference 219 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RoboClaw · Reference 220 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RLBench · Reference 227 · evaluation · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • LIBERO · Reference 228 · evaluation · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RoboTwin 2.0 · Reference 230 · evaluation · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Wall-OSS-0.5 · Reference 233 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Past-Token Prediction (PTP) · Reference 235 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • RoboAgent / MT-ACT · Reference 243 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • XR-2 · Reference 244 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • MimicDroid · Reference 245 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Xiaomi-Robotics-1 · Reference 247 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • MimicPlay · Reference 251 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • TecNets: Humans without Humans · Reference 260 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • SUN Programs / Kuafu · Reference 262 · agent · Cited source ↗

    Cited as: SUN: Persistent Programs For Language-Grounded Control-to-Learning-to-Real Policies

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RoboCasa365 · Reference 263 · data · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Dream2Fix · Reference 270 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • EXIMO · Reference 272 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • π*0.6 / RECAP · Reference 273 · policy · Cited source ↗

    Cited as: π^*_0.6: a VLA That Learns From Experience

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • ACT · Reference 274 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Octo · Reference 275 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RT-1 · Reference 276 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • RT-2 · Reference 277 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Heterogeneous Pre-trained Transformers (HPT) · Reference 278 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • RDT-1B · Reference 279 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Human-to-Robot VLA Transfer · Reference 280 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • π0.7 · Reference 281 · policy · Cited source ↗

    Cited as: π_0.7: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities

    The arXiv paper is the existing π0.7 work; its paper PDF URL previously prevented an exact URL match.

  • LipVQ-VAE / ICRT · Reference 282 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • HiST-AT · Reference 283 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • MG-VQA Agent · Reference 291 · agent · Cited source ↗

    Cited as: PROBE: Manipulation-Grounded Visual Question Answering with VLM Agents

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • CR-VLA-Force / CC-VLA · Reference 292 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • ST-GPT / SGPT · Reference 293 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Visual MPC with Retrying · Reference 294 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • ScrewMimic · Reference 295 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • PETNet · Reference 297 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Introspective Planning · Reference 298 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • TacForcing · Reference 299 · policy · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • MetaCURE · Reference 300 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • CoRe · Reference 302 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • AgentChord · Reference 303 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • VoLoAgent · Reference 304 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • ICLHF · Reference 305 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • HoloMind / LongAct · Reference 307 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • LIBERO-Plus · Reference 310 · evaluation · Cited source ↗

    Cited as: LIBERO-Plus: In-depth Robustness Analysis of Vision-Language-Action Models

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RoboMonkey · Reference 315 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • τ0-VLA · Reference 316 · world · Cited source ↗

    Cited as: τ_0-VLA: a Hierarchical Robot Foundation Model with World-Model-Guided Test-Time Computation

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • SAIL · Reference 317 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Fast-ThinkAct · Reference 325 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • SILVR · Reference 327 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • VLAW · Reference 328 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • World-VLA-Loop · Reference 329 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Motus2 · Reference 330 · world · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Behavioral Exploration · Reference 333 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • GAVEL · Reference 336 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • StageGuard · Reference 337 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • LIFD · Reference 339 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • SkipVLA · Reference 340 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Closed-Loop Robot Software · Reference 341 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • HIL-UMI · Reference 342 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • VA-Bench Shared Agent Harness · Reference 344 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • REVOLVE · Reference 345 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Relational NDF · Reference 346 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • Parts-based Shape Warping · Reference 347 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • RoboRSI (Noematrix) · Reference 348 · agent · Cited source ↗

    Cited as: RoboRSI: Stable, Efficient, and Reusable Robot Self-Evolution in Complex Real-World Environments

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • OpenETA · Reference 349 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RoboHarness · Reference 350 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Zero-Shot Visual Imitation / GSP · Reference 355 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Zeva-Ego · Reference 363 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • RopeFormer · Reference 364 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Robo Dawn · Reference 365 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Sandwich-Residuals · Reference 367 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • MetaPusher · Reference 368 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • Weights or Skills? · Reference 370 · foundations · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • World-Action Models — Learning and Control · Reference 371 · foundations · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Demonstration-driven Manipulation ICL · Reference 372 · foundations · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Robo-Harness K1 · Reference 379 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RACaP · Reference 380 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • MemBodied · Reference 381 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • EmbodiedSWE · Reference 382 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • TP-Flow · Reference 386 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • ARMS · Reference 387 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • World Action Agent (WAA) · Reference 388 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • GLOW · Reference 389 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • RAPID · Reference 390 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Behavior Predictive Control (BPC) · Reference 391 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • TraceFlow · Reference 392 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Self-Adaptive VLA · Reference 393 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • Mimir · Reference 394 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • OCC4M · Reference 395 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • Workspace Models · Reference 397 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • HarnessPAI · Reference 398 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • MaP-WAM · Reference 401 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • FaRe · Reference 402 · world · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • 2AM · Reference 403 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • HIRE · Reference 404 · policy · Cited source ↗

    Verified manipulation policy in the original primary source.

  • POIL · Reference 405 · geometry · Cited source ↗

    Original primary source verifies a manipulation policy, planner or controller and a concrete robot-manipulation evaluation.

  • Coding Agents for Generalized TAMP · Reference 407 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • RIVET · Reference 408 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • HuGo · Reference 409 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

  • DualManip · Reference 410 · agent · Cited source ↗

    Linked to a canonical atlas record. Technical fields are supported by that work's own primary sources.

Outside manipulation scope · 33
Background reference · 37
Adjacent data resource · 42
Candidate for a separate audit · 61

Tactile-Guided Robot Learning: From Representation Learning to Foundation Models

Complete original bibliography with explicit identity matches and individually reviewed title-level dispositions. Indexed means a reviewed canonical record exists; background and adjacent resources are not counted as method reviews. Candidate means original-source review is still pending. No fuzzy identity matching or containing-survey fallback matching is used. Snapshot: Original bibliography, manual dispositions and exact original-source late-batch reconciliation, 2026-10-17.

Author reference inventory ↗ · Author method corpus ↗

Indexed in this atlas · 72
  • RoboPack · Reference 3 · policy · Cited source ↗

    Cited as: Robopack: learning tactile-informed dynamics models for dense packing

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • TacSL · Reference 4 · simulator · Cited source ↗

    Cited as: Tacsl: a library for visuotactile sensor simulation and learning

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Compliant-Hand Visuotactile Insertion · Reference 7 · policy · Cited source ↗

    Cited as: Visuotactile-based learning for insertion with compliant hands

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • VLA-Touch · Reference 13 · policy · Cited source ↗

    Cited as: Vla-touch: enhancing vision-language-action model with dual-level tactile feedback

    Manually verified original-source alias: Survey uses singular model and a published-year variation; original arXiv title and the same Bi/Ma/Hao/Shou/Soh authors verify VLA-Touch identity.

  • π0 · Reference 16 · policy · Cited source ↗

    Cited as: p ​ i ​ _ ​ 0 pi\_0 : A vision-language-action flow model for general robot control

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Tactile Multimodal Fusion — Survey · Reference 20 · survey · Cited source ↗

    Cited as: Tactile-based multimodal fusion in embodied intelligence: a survey of vision, language, and contact-driven paradigms

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • UniVTAC · Reference 23 · benchmark · Cited source ↗

    Cited as: Univtac: a unified simulation platform for visuo-tactile manipulation data generation, learning, and benchmarking

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • ImplicitRDP · Reference 25 · policy · Cited source ↗

    Cited as: ImplicitRDP: an end-to-end visual-force diffusion policy with structural slow-fast learning

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • UniForce · Reference 26 · model · Cited source ↗

    Cited as: UniForce: a unified latent force model for robot manipulation with diverse tactile sensors

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • TLV-Link / Touch100k · Reference 28 · model · Cited source ↗

    Cited as: Touch100k: a large-scale touch-language-vision dataset for touch-centric multimodal representation

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • OmniVTLA · Reference 29 · policy · Cited source ↗

    Cited as: OmniVTLA: vision-tactile-language-action models with semantic-aligned tactile sensing

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Diffusion Policy · Reference 30 · policy · Cited source ↗

    Cited as: Diffusion policy: visuomotor policy learning via action diffusion

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • MViTac · Reference 36 · model · Cited source ↗

    Cited as: Multimodal visual-tactile representation learning through self-supervised contrastive pre-training

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • AnyTouch · Reference 41 · model · Cited source ↗

    Cited as: Anytouch: learning unified static-dynamic representation across multiple visuo-tactile sensors

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • AnyTouch 2 / ToucHD · Reference 42 · model · Cited source ↗

    Cited as: Anytouch 2: general optical tactile representation learning for dynamic tactile perception

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Text2Touch · Reference 44 · policy · Cited source ↗

    Cited as: Text2Touch: tactile in-hand manipulation with llm-designed reward functions

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • TVL / TVL-LLaMA · Reference 45 · model · Cited source ↗

    Cited as: A touch, vision, and language dataset for multimodal alignment

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • TVL / TVL-LLaMA · Reference 46 · model · Cited source ↗

    Cited as: A touch, vision, and language dataset for multimodal alignment

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • VITaL Pretraining · Reference 49 · policy · Cited source ↗

    Cited as: VITaL pretraining: visuo-tactile pretraining for tactile and non-tactile manipulation policies

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • SITR · Reference 55 · model · Cited source ↗

    Cited as: Sensor-invariant tactile representation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • T-Dex · Reference 56 · policy · Cited source ↗

    Cited as: Dexterity from touch: self-supervised pre-training of tactile representations with robotic play

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • TLA · Reference 58 · policy · Cited source ↗

    Cited as: TLA: tactile-language-action model for contact-rich manipulation

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by exact normalized original title. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • FoAR · Reference 60 · policy · Cited source ↗

    Cited as: Foar: force-aware reactive policy for contact-rich robotic manipulation

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by exact normalized original title. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • ViTacFormer · Reference 61 · policy · Cited source ↗

    Cited as: ViTacFormer: learning cross-modal representation for visuo-tactile dexterous manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Sparsh / TacBench · Reference 62 · model · Cited source ↗

    Cited as: Sparsh: self-supervised touch representations for vision-based tactile sensing

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Sparsh-X · Reference 63 · model · Cited source ↗

    Cited as: Tactile beyond pixels: multisensory touch representations for robot manipulation

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • 3D-ViTac · Reference 68 · policy · Cited source ↗

    Cited as: 3D-vitac: learning fine-grained manipulation with visuo-tactile sensing

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • VT-Refine · Reference 69 · policy · Cited source ↗

    Cited as: VT-refine: learning bimanual assembly with visuo-tactile feedback via simulation fine-tuning

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Tactile-VLA · Reference 70 · policy · Cited source ↗

    Cited as: Tactile-vla: unlocking vision-language-action model’s physical knowledge for tactile generalization

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • ViTacWorld · Reference 73 · model · Cited source ↗

    Cited as: ViTacWorld: scaling visuo-tactile world models for contact-rich robot manipulation

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • TaF-VLA · Reference 74 · policy · Cited source ↗

    Cited as: Tactile-force alignment in vision-language-action models for force-aware manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • π0.5 · Reference 76 · policy · Cited source ↗

    Cited as: p ​ i ​ _ ​ 0.5 pi\_{0.5} : A vision-language-action model with open-world generalization

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • BC-Z · Reference 79 · policy · Cited source ↗

    Cited as: Bc-z: zero-shot task generalization with robotic imitation learning

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Making Sense of Vision and Touch · Reference 88 · policy · Cited source ↗

    Cited as: Making sense of vision and touch: self-supervised learning of multimodal representations for contact-rich tasks

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • Tac-DINO / VTPA · Reference 91 · model · Cited source ↗

    Cited as: Tac-dino: learning vision-tactile features with patch alignment

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • AdapTac · Reference 92 · policy · Cited source ↗

    Cited as: Adaptive visuo-tactile fusion with predictive force attention for dexterous manipulation

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • Tactile Functasets / TactFunc · Reference 96 · model · Cited source ↗

    Cited as: Tactile functasets: neural implicit representations of tactile datasets

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • PP-Tac · Reference 98 · policy · Cited source ↗

    Cited as: PP-tac: paper picking using omnidirectional tactile feedback in dexterous robotic hands

    Manually verified original-source alias: Workshop citation inserts omnidirectional into the title; the same six Lin/Huang/Li/Ma/Xiao/Jiao authors and PP-Tac method verify the original paper identity.

  • ViTaMIn · Reference 100 · policy · Cited source ↗

    Cited as: Vitamin: learning contact-rich tasks through robot-free visuo-tactile manipulation interface

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • TacFiLM · Reference 115 · policy · Cited source ↗

    Cited as: Tactile modality fusion for vision-language-action models

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Open X-Embodiment / RT-X · Reference 125 · policy · Cited source ↗

    Cited as: Open x-embodiment: robotic learning datasets and rt-x models: open x-embodiment collaboration 0

    Manually verified original-source alias: Bibliography appends Open X-Embodiment Collaboration text to the original title; O’Neill and collaborators match the original RT-X paper.

  • TactX · Reference 126 · model · Cited source ↗

    Cited as: TactX: learning shared tactile representations across diverse sensors

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • CTTP / Contrastive Touch-to-Touch Pretraining · Reference 131 · model · Cited source ↗

    Cited as: Contrastive touch-to-touch pretraining

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • CTTP / Contrastive Touch-to-Touch Pretraining · Reference 132 · model · Cited source ↗

    Cited as: Contrastive touch-to-touch pretraining

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • M3L — Power of the Senses · Reference 136 · policy · Cited source ↗

    Cited as: The power of the senses: generalizable manipulation from vision and touch through masked multimodal learning

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • ManiSkill3 · Reference 140 · simulator · Cited source ↗

    Cited as: Maniskill3: gpu parallelized robotics simulation and rendering for generalizable embodied ai

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • TacVerse · Reference 150 · dataset · Cited source ↗

    Cited as: TacVerse: a multi-sensor dataset and benchmark for cross-sensor vision-based tactile perception

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • DexVLA · Reference 151 · policy · Cited source ↗

    Cited as: Dexvla: vision-language model with plug-in diffusion expert for general robot control

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • FreeTacMan · Reference 154 · policy · Cited source ↗

    Cited as: FreeTacMan: robot-free visuo-tactile data collection system for contact-rich manipulation

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Canonical 3D Tactile Pretraining · Reference 155 · policy · Cited source ↗

    Cited as: Canonical representation and force-based pretraining of 3d tactile for dexterous visuo-tactile policy learning

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • TacDiffusion · Reference 156 · policy · Cited source ↗

    Cited as: Tacdiffusion: force-domain diffusion policy for precise tactile manipulation

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Forceful Robotic Foundation Models — Survey · Reference 157 · survey · Cited source ↗

    Cited as: Towards forceful robotic foundation models: a literature survey

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • LeTac-MPC · Reference 161 · policy · Cited source ↗

    Cited as: LeTac-mpc: learning model predictive control for tactile-reactive grasping

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • UniT · Reference 162 · model · Cited source ↗

    Cited as: UniT: data efficient tactile representation with generalization to unseen objects

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Reactive Diffusion Policy · Reference 163 · policy · Cited source ↗

    Cited as: Reactive diffusion policy: slow-fast visual-tactile policy learning for contact-rich manipulation

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • UniTouch / Touch-LLM · Reference 165 · model · Cited source ↗

    Cited as: Binding touch to everything: learning unified multimodal tactile representations

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Touch and Go · Reference 166 · model · Cited source ↗

    Cited as: Touch and go: learning from human-collected vision and touch

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by exact normalized original title. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • Tactile Skin Translation · Reference 169 · policy · Cited source ↗

    Cited as: Learning in-hand translation using tactile skin with shear and normal force sensing

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Octopi-1.5 · Reference 173 · model · Cited source ↗

    Cited as: Demonstrating the octopi-1.5 visual-tactile-language model

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Octopi / PhysiCLeAR · Reference 174 · model · Cited source ↗

    Cited as: Octopi: object property reasoning with large tactile-language models

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • FTP-1 · Reference 175 · policy · Cited source ↗

    Cited as: FTP-1: a generalist foundation tactile policy across tactile sensors for contact-rich manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • VTLA · Reference 178 · policy · Cited source ↗

    Cited as: Vtla: vision-tactile-language-action model with preference learning for insertion manipulation

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by exact normalized original title. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • TacVLA · Reference 180 · policy · Cited source ↗

    Cited as: Tacvla: contact-aware tactile fusion for robust vision-language-action manipulation

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • ReTouch · Reference 181 · policy · Cited source ↗

    Cited as: ReTouch: empowering contact-rich dexterous manipulation with online-refined tactile prediction

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • ContactWorld · Reference 183 · benchmark · Cited source ↗

    Cited as: ContactWorld: what matters in vision-tactile world models for contact-rich manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • T3 / Foundation Tactile · Reference 185 · model · Cited source ↗

    Cited as: Transferable tactile transformers for representation learning across diverse sensors and tasks

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • ACT · Reference 186 · policy · Cited source ↗

    Cited as: Learning fine-grained bimanual manipulation with low-cost hardware

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • ViTac-Tracing · Reference 187 · policy · Cited source ↗

    Cited as: Vitac-tracing: visual-tactile imitation learning of deformable object tracing

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • ViTaL — local skills · Reference 188 · policy · Cited source ↗

    Cited as: Touch begins where vision ends: generalizable policies for contact-rich manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Tac-Man · Reference 190 · policy · Cited source ↗

    Cited as: Tac-Man: tactile-informed prior-free manipulation of articulated objects

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • TacMan-Turbo · Reference 191 · policy · Cited source ↗

    Cited as: TacMan-Turbo: proactive tactile control for robust and efficient articulated object manipulation

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by exact normalized original title. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • TouchWorld · Reference 192 · policy · Cited source ↗

    Cited as: TouchWorld: a predictive and reactive tactile foundation model for dexterous manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

Background reference · 45
Adjacent data resource · 39
Candidate for a separate audit · 39

Tactile-based Multimodal Fusion in Embodied Intelligence: A Survey of Vision, Language, and Contact-Driven Paradigms

Complete original bibliography with explicit identity matches and individually reviewed title-level dispositions. Indexed means a reviewed canonical record exists; background and adjacent resources are not counted as method reviews. Candidate means original-source review is still pending. No fuzzy identity matching or containing-survey fallback matching is used. Snapshot: Original bibliography, manual dispositions and exact original-source late-batch reconciliation, 2026-10-17.

Author reference inventory ↗ · Author method corpus ↗

Indexed in this atlas · 28
  • MViTac · Reference 1 · model · Cited source ↗

    Cited as: Multimodal visual-tactile representation learning through self-supervised contrastive pre-training

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • UniTouch / Touch-LLM · Reference 2 · model · Cited source ↗

    Cited as: Binding touch to everything: learning unified multimodal tactile representations

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • TVL / TVL-LLaMA · Reference 3 · model · Cited source ↗

    Cited as: A touch, vision, and language dataset for multimodal alignment

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • CLTP / Tac3D-LLM · Reference 5 · model · Cited source ↗

    Cited as: CLTP: contrastive language-tactile pre-training for 3d contact geometry understanding

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • TLV-Link / Touch100k · Reference 6 · model · Cited source ↗

    Cited as: Touch100k: a large-scale touch-language-vision dataset for touch-centric multimodal representation

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • VTLA · Reference 7 · policy · Cited source ↗

    Cited as: Vtla: vision-tactile-language-action model with preference learning for insertion manipulation

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • VTV-LLM / VTV150K · Reference 8 · model · Cited source ↗

    Cited as: Universal visuo-tactile video understanding for embodied interaction

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • AnyTouch · Reference 9 · model · Cited source ↗

    Cited as: Anytouch: learning unified static-dynamic representation across multiple visuo-tactile sensors

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • RA-Touch · Reference 13 · model · Cited source ↗

    Cited as: RA-touch: retrieval-augmented touch understanding with enriched visual data

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by exact normalized original title. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • More Than a Feeling · Reference 43 · policy · Cited source ↗

    Cited as: More than a feeling: learning to grasp and regrasp using vision and touch

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Touch and Go · Reference 46 · model · Cited source ↗

    Cited as: Touch and go: learning from human-collected vision and touch

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • Octopi / PhysiCLeAR · Reference 50 · model · Cited source ↗

    Cited as: Octopi: object property reasoning with large tactile-language models

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • SToLa · Reference 51 · model · Cited source ↗

    Cited as: SToLa: self-adaptive touch-language framework with tactile commonsense reasoning in open-ended scenarios

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • OmniVTLA · Reference 55 · policy · Cited source ↗

    Cited as: Omnivtla: vision-tactile-language-action model with semantic-aligned tactile sensing

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • TLA · Reference 59 · policy · Cited source ↗

    Cited as: Tla: tactile-language-action model for contact-rich manipulation

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • FreeTacMan · Reference 60 · policy · Cited source ↗

    Cited as: Freetacman: robot-free visuo-tactile data collection system for contact-rich manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • OmniVTA · Reference 64 · policy · Cited source ↗

    Cited as: OmniVTA: visuo-tactile world modeling for contact-rich robotic manipulation

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • Making Sense of Vision and Touch · Reference 71 · policy · Cited source ↗

    Cited as: Making sense of vision and touch: self-supervised learning of multimodal representations for contact-rich tasks

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by exact normalized original title. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • Visuo-Tactile Transformers · Reference 75 · policy · Cited source ↗

    Cited as: Visuo-tactile transformers for manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Octopi-1.5 · Reference 83 · model · Cited source ↗

    Cited as: Demonstrating the octopi-1.5 visual-tactile-language model

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • ViTacFormer · Reference 85 · policy · Cited source ↗

    Cited as: ViTacFormer: learning cross-modal representation for visuo-tactile dexterous manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • ViTaPEs · Reference 86 · model · Cited source ↗

    Cited as: ViTaPEs: visuotactile position encodings for cross-modal alignment in multimodal transformers

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • ConViTac · Reference 91 · model · Cited source ↗

    Cited as: ConViTac: aligning visual-tactile fusion with contrastive representations

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • DreamTacVLA · Reference 95 · policy · Cited source ↗

    Cited as: Learning to feel the future: dreamtacvla for contact-rich manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • VLA-Touch · Reference 96 · policy · Cited source ↗

    Cited as: Vla-touch: enhancing vision-language-action models with dual-level tactile feedback

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • VTAM · Reference 99 · policy · Cited source ↗

    Cited as: VTAM: video-tactile-action models for complex physical interaction beyond vlas

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • TacThru-UMI · Reference 104 · policy · Cited source ↗

    Cited as: Simultaneous tactile-visual perception for learning multimodal robot manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Making Sense of Vision and Touch · Reference 119 · policy · Cited source ↗

    Cited as: Making sense of vision and touch: learning multimodal representations for contact-rich tasks

    Manually verified journal-extension identity: the 2020 T-RO title and nine named authors match original arXiv:1907.13098, which the reviewed making-sense-vision-touch record explicitly links as its extended manuscript. Quantitative fields remain version-qualified to the original ICRA paper.

Background reference · 25
  • A survey of deep learning and its applications: a new paradigm to machine learning ↗ · Reference 14 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Attention is all you need ↗ · Reference 15 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • A survey of large language models ↗ · Reference 16 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Imagebind: one embedding space to bind them all ↗ · Reference 17 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Deep residual learning for image recognition ↗ · Reference 20 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • An image is worth 16x16 words: transformers for image recognition at scale ↗ · Reference 21 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Bert: pre-training of deep bidirectional transformers for language understanding ↗ · Reference 22 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Learning transferable visual models from natural language supervision ↗ · Reference 23 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Multimodal alignment and fusion: a survey ↗ · Reference 37 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • A survey on multimodal large language models ↗ · Reference 38 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Imagenet classification with deep convolutional neural networks ↗ · Reference 65 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Mask r-cnn ↗ · Reference 67 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Bayesian neural networks ↗ · Reference 68 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Generative adversarial nets ↗ · Reference 70 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Flownet: learning optical flow with convolutional networks ↗ · Reference 72 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Very deep convolutional networks for large-scale image recognition ↗ · Reference 74 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Mastering visual continuous control: improved data-augmented reinforcement learning ↗ · Reference 77 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Mlp-mixer: an all-mlp architecture for vision ↗ · Reference 79 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Fine-tuned clip models are efficient video learners ↗ · Reference 80 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Vicuna: an open-source chatbot impressing gpt-4 with 90%* chatgpt quality ↗ · Reference 81 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Qwen2-vl: enhancing vision-language model’s perception of the world at any resolution ↗ · Reference 84 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Efficientnetv2: smaller models and faster training ↗ · Reference 89 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Ulip-2: towards scalable multimodal pre-training for 3d understanding ↗ · Reference 90 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Densely connected convolutional networks ↗ · Reference 94 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

  • Gpt-4o system card ↗ · Reference 97 · reference

    Foundational machine learning, general control theory, human sensory biology, or general contextual reference; retained as background, not a separately reviewed RGBDT benchmark or manipulation method.

Adjacent data resource · 48
Candidate for a separate audit · 18

Towards Forceful Robotic Foundation Models: a Literature Survey

Complete original bibliography with explicit identity matches and individually reviewed title-level dispositions. Indexed means a reviewed canonical record exists; background and adjacent resources are not counted as method reviews. Candidate means original-source review is still pending. No fuzzy identity matching or containing-survey fallback matching is used. Snapshot: Original bibliography, manual dispositions and exact original-source late-batch reconciliation, 2026-10-17.

Author reference inventory ↗ · Author method corpus ↗

Indexed in this atlas · 25
  • RT-1 · Reference 2 · policy · Cited source ↗

    Cited as: Rt-1: Robotics transformer for real-world control at scale

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Open X-Embodiment / RT-X · Reference 3 · policy · Cited source ↗

    Cited as: Open X-Embodiment: Robotic learning datasets and RT-X models

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • OpenVLA · Reference 4 · policy · Cited source ↗

    Cited as: Openvla: An open-source vision-language-action model

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • π0 · Reference 5 · policy · Cited source ↗

    Cited as: Pi0: Our first generalist policy

    Manually verified original-source alias: The citation is the official PI blog Our First Generalist Policy, already the canonical pi0 project link; the paper and blog are the same policy release.

  • Diffusion Policy · Reference 11 · policy · Cited source ↗

    Cited as: Diffusion policy: Visuomotor policy learning via action diffusion

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • SayCan · Reference 41 · policy · Cited source ↗

    Cited as: Do as i can, not as i say: Grounding language in robotic affordances

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • BC-Z · Reference 42 · policy · Cited source ↗

    Cited as: Bc-z: Zero-shot task generalization with robotic imitation learning

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • OpenVLA · Reference 44 · policy · Cited source ↗

    Cited as: Openvla: An open-source vision-language-action model

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • TinyVLA · Reference 45 · policy · Cited source ↗

    Cited as: Tinyvla: Towards fast, data-efficient vision-language-action models for robotic manipulation

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • ACT · Reference 47 · policy · Cited source ↗

    Cited as: Learning fine-grained bimanual manipulation with low-cost hardware

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Making Sense of Vision and Touch · Reference 62 · policy · Cited source ↗

    Cited as: Making Sense of Vision and Touch: Self-Supervised Learning of Multimodal Representations for Contact-Rich Tasks, Mar

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • ViSk · Reference 64 · policy · Cited source ↗

    Cited as: Learning Precise, Contact-Rich Manipulation through Uncalibrated Tactile Skins, Oct

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • FACTR · Reference 66 · policy · Cited source ↗

    Cited as: Factr: Force-attending curriculum training for contact-rich policy learning

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Octo · Reference 71 · policy · Cited source ↗

    Cited as: Octo: An open-source generalist robot policy

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • TinyVLA · Reference 73 · policy · Cited source ↗

    Cited as: Tinyvla: Towards fast, data-efficient vision-language-action models for robotic manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • FoAR · Reference 77 · policy · Cited source ↗

    Cited as: FoAR: Force-Aware Reactive Policy for Contact-Rich Robotic Manipulation, Nov

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • 3D-ViTac · Reference 79 · policy · Cited source ↗

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

  • FuSe · Reference 80 · policy · Cited source ↗

    Cited as: Beyond Sight: Finetuning Generalist Robot Policies with Heterogeneous Sensors via Language Grounding, Jan

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • ForceMimic / HybridIL · Reference 84 · policy · Cited source ↗

    Cited as: ForceMimic: Force-Centric Imitation Learning with Force-Motion Capture System for Contact-Rich Manipulation, Oct. 2024b

    Original-source identity reconciled to the independently reviewed canonical/late-batch record by original arXiv identifier. Scientific details and release scope are supported by that record; the containing bibliography URL is provenance only.

  • TacDiffusion · Reference 89 · policy · Cited source ↗

    Cited as: TacDiffusion: Force-domain Diffusion Policy for Precise Tactile Manipulation, Sept

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • UniT · Reference 90 · model · Cited source ↗

    Cited as: UniT: Unified Tactile Representation for Robot Learning, Aug

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • MimicTouch · Reference 91 · policy · Cited source ↗

    Cited as: MimicTouch: Leveraging Multi-modal Human Tactile Demonstrations for Contact-rich Manipulation, Sept

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • T3 / Foundation Tactile · Reference 92 · model · Cited source ↗

    Cited as: Transferable Tactile Transformers for Representation Learning Across Diverse Sensors and Tasks, Oct

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • Reactive Diffusion Policy · Reference 95 · policy · Cited source ↗

    Cited as: Reactive diffusion policy: Slow-fast visual-tactile policy learning for contact-rich manipulation

    Matched to a reviewed canonical record by its original arXiv identifier; model/policy specifications retain independent original-source evidence.

  • TVL / TVL-LLaMA · Reference 100 · model · Cited source ↗

    Canonical original work matched by arXiv identity or normalized title; original model/policy evidence is retained in its own record.

Background reference · 32
Adjacent data resource · 17
Candidate for a separate audit · 35

Download the survey reference mappings ↗

What the papers disclose

The requested cost and implementation fields are recorded for every policy. Some authors disclose only selected settings, so populated values retain their component and experiment qualifiers.

Policy fieldReported / describedUndisclosed / unresolved
Model size238368
Training data58917
Training duration300306
Training hardware258348
Inference / control rate293313

Search and review method

Coverage limits

This collection focuses on manipulation and action-policy evaluation. Tactile representations and manipulation-relevant depth/geometry models have dedicated routes. Standalone sensor hardware, medical palpation, locomotion, haptics and generic depth-estimation papers are not exhaustively cataloged. Selected supporting resources are included when they explain the benchmarks or policies.

A listed task registry can contain more tasks than a paper’s evaluation subset. Task lists are labeled as complete suites, released registries, subsets or examples. Task totals across overlapping suites and datasets are not aggregated.

Recent anonymous or unreleased projects are included only when an accessible author source supports the record. Their availability is qualified; an unpublished checkpoint or an undocumented simulator version is never assumed.

Update the local repository

The structured source is data/works.json; figure provenance is data/media.json. Edit a record with primary-source evidence, then run:

python3 scripts/generate.py
python3 scripts/validate.py
cd website && npm test && npm run check && npm run build

Review additions in the local browser before release. Validated main-branch changes deploy to the public GitHub Pages site. Download the complete structured catalog ↗