Flow Matching Imitation Learning for Multi-Support Manipulation

Humanoid robots could benefit from using their upper bodies for supportcontacts, enhancing their workspace, stability, and ability to performcontact-rich and pushing tasks. In this paper, we propose a unified approachthat combines an optimization-based multi-contact whole-body controller withFlow Matching, a recently introduced method capable of generating multi-modaltrajectory distributions for imitation learning. In simulation, we show thatFlow Matching is more appropriate for robotics than Diffusion and traditionalbehavior cloning. On a real full-size humanoid robot (Talos), we demonstratethat our approach can learn a whole-body non-prehensile box-pushing task andthat the robot can close dishwasher drawers by adding contacts with its freehand when needed for balance. We also introduce a shared autonomy mode forassisted teleoperation, providing automatic contact placement for tasks notcovered in the demonstrations. Full experimental videos are available at:https://hucebot.github.io/flow_multisupport_website/

Further reading#

• Access Paper in arXiv.org