RoboGen

Why we included this project

RoboGen is a research codebase that runs the whole robot skill acquisition loop on its own: it comes up with a manipulation or locomotion task, builds a matching scene in a physics simulator, and trains a policy to solve it, with no human in the middle. For teams working on embodied AI or robot learning, it's a useful reference for wiring a large language model to a physics engine and motion planning so you can generate training data in bulk. The repo includes a working PyBullet implementation for rigid-body manipulation and locomotion, plus example configs and scripts that turn a text description into a task or train skills with RL. Treat it as a research artifact to study and adapt rather than something you'd drop into production: task generation depends on external assets like PartNet-Mobility and an OpenAI API key.