ManipGen

ManipGen is a generalist agent for long-horizon robotic manipulation. It leverages a new class of policies for sim2real transfer: local policies. Locality enables a variety of appealing properties including invariances to absolute robot and object pose, skill ordering, and global scene configuration. We combine these policies with foundation models for vision, language and motion planning to solve a wide range of long-horizon manipulation tasks in the real world zero-shot.

All ManipGen checkpoints, as well as our codebase are released under an MIT License.

For full details, please read our paper and see our project page.

Model Summary

• Developed by: The ManipGen team consisting of researchers from Carnegie Mellon University and Apple.
• License: MIT
• Repository: https://github.com/mihdalal/manipgen
• Paper: https://arxiv.org/abs/2410.22332
• Project website & videos: https://mihdalal.github.io/manipgen/