Human skill. Robot potential.

A living collection of interfaces, datasets, and ideas for robot learning.

75 hardware entries

HKUST · 2026

FOMI

A scissor-shaped UMI-style interface provides direct mechanical force feedback.

VisionProprioception

Daimon Robotics · 2026

TacClaw-Infinity

A visuotactile collection interface supporting the TacClaw-Infinity dataset.

VisionTactileIMU

Stanford · 2026

PanTact

Wearable contact patches and UMI grippers for whole-body contact demonstrations.

VisionTactileContact

Zhejiang University · 2026

VOMMI

Chest and wrist RGB views provide portable mobile-manipulation demonstrations.

VisionProprioception

Northwestern University · 2026

PolyUMI

A wireless collector combining vision, touch, and contact audio in one sensing finger.

VisionTactileAudio

Autel US · 2026

KIWI

Camera-only bimanual capture, with both hands registered in one metric map.

VisionIMUAudio

Zhejiang University · 2026

Whole-Body UMI

Native UMI end-effector skills drive humanoid whole-body motion generation.

VisionProprioception

Peking University · 2026

SEED-UMI

A shared 20-DoF exoskeleton for human demonstration and dexterous robot deployment.

VisionProprioception

NeoteAI · 2026

N0-Foundation / TacUMI

An infrared-tracked gripper combining RGB vision and two tactile sensors.

VisionTactileProprioception

Huaweike Intelligent Tech / 华威科 · 2026

HIT UMI Tactile Collection

A manufacturer UMI solution combining tracked hand motion with tactile acquisition.

VisionTactileProprioception

Simple AI · 2026

HiFi-UMI

High-fidelity, portable capture with stereo-inertial tracking and a released 2,000-hour dataset.

VisionIMUProprioception

X Square Robot · 2026

QUANXTA Zero-G1

A headband and dual-gripper collector with synchronized vision, tactile and audio.

VisionTactileAudio

A clearer view of UMI.

Awesome UMI connects the hardware, datasets, and research behind robot learning from human demonstrations.

The README is the canonical collection. Specifications and release information are checked against papers and official resources. Images belong to their credited projects.

Paper-reported scale, indexed data, and downloadable releases are kept separate. Missing information stays unverified.

Read the collection ↗