Back to Gap Map Model components / Robotics & embodied AI

robomimic

ARISE Initiative
open source / Overall score: 2.0

robomimic is a framework for robot learning from demonstration. It pairs released datasets of human and machine-generated manipulation demonstrations with imitation-learning and offline reinforcement learning algorithms, including BC-RNN, BC-Transformer, IQL and Diffusion Policy, and with tools for configuring, sweeping and reproducing training runs. It is part of the ARISE Initiative and uses robosuite as its simulator.

Openness

5 high confidence
5.0
license
MIT (OSI, released datasets also MIT on the Hub)
source
public
core features withheld
no — an academic project with no paid offering

The framework is MIT-licensed and builds in full from its public repository, and the demonstration datasets it distributes carry the same license. It is an academic project begun at Stanford and maintained across several labs, with no paid edition.

Adoption

2 medium confidence
2.0

PyPI downloads of the robomimic package. The newest release on PyPI is older than the newest GitHub release, and the documentation recommends installing from source, so the count misses users of the current version.

Capability

not assessed

The software in this category is compared on how fast and how faithfully it simulates robots. robomimic trains policies from recorded demonstrations and relies on robosuite for any simulation, so it has no place on that comparison and none is claimed.

Verified 2026-11-12