Keywords: Sim-to-real Transfer, Long-horizon Manipulation
TL;DR: Introduces a new class of local policies for sim2real transfer in robotic manipulation, achieving state-of-the-art zero-shot performance on diverse long-horizon tasks.
Abstract: Sim2real for robotic manipulation is difficult due to the challenges of simulating complex contacts and generating realistic task distributions. To tackle the latter problem, we introduce ManipGen, which 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 and demonstrate SOTA zero-shot performance of our method to Robosuite benchmark tasks in simulation (97%). We transfer our local policies from simulation to reality and observe they can solve unseen long-horizon manipulation tasks with up to 8 stages with significant pose, object and scene configuration variation. ManipGen outperforms SOTA approaches such as SayCan, OpenVLA and LLMTrajGen across 50 real-world manipulation tasks by 36%, 76% and 62% respectively. All code, models and datasets will be released. Video results at https://manipgen.github.io/.
Submission Number: 23
Loading