real-stanford/universal_manipulation_interface

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1560 stars · 285 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Universal Manipulation Interface (UMI) is a data collection and policy learning framework that transfers in-the-wild human demonstrations into deployable robot manipulation policies using handheld grippers. It includes a SLAM pipeline for processing demonstrations and a policy interface with latency matching and relative-trajectory action representation for hardware-agnostic deployment.

Use cases

• collect robot manipulation demonstrations with handheld grippers
• train robot policies from human demonstrations
• run SLAM on gripper-mounted camera footage
• deploy manipulation policies across different robot hardware
• learn bimanual and dynamic manipulation skills
• zero-shot generalize manipulation policies to new environments

When to choose

• you need low-cost portable robot demonstration data collection without a robot
• you want hardware-agnostic manipulation policies trained on human demos
• you need bimanual, precise, or long-horizon manipulation capabilities

When to avoid

• you need a general-purpose robotics simulator
• you work outside Ubuntu 22.04 and cannot use Docker
• you need locomotion or mobile navigation rather than manipulation

Facets

framework · maturity active

machine-learning robotics computer-vision data-science robotics machine-learning artificial-intelligence python robot-manipulation imitation-learning slam handheld-gripper policy-learning demonstration-collection linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="real-stanford/universal_manipulation_interface")