FLORE learns force-reactive humanoid loco-manipulation from handheld demos
Original title: Learning Force-Reactive Humanoid Loco-Manipulation
On three contact-rich robot tasks the force-reactive policy succeeds in 80% of rollouts versus 26.7% for position-only.
Researchers present FLORE, a framework that learns coordinated hand force and whole-body motion for humanoids. Robot-free demonstrations use two force-sensing UMI-style grippers and wearable body trackers. A diffusion policy predicts actions and per-hand force commands, and a hybrid compliance controller tracks force while balancing. In simulation, force commands of varied magnitude and direction track with little loss in motion accuracy.
- Force tracking uses an integral of error from real force/torque sensors.
- Each hand tracks force along the commanded direction and stays compliant otherwise.
- The position-only baseline used the same demonstrations and the same controller.
- Simulation checks per-hand force commands of varied magnitude and direction.
From the original article
1 min read at arxiv.orgMany everyday tasks require controlling contact force as well as motion. On a humanoid, this force must often be coordinated across both hands and the whole body, and cannot be specified by kinematic trajectories alone. Therefore, we present FLORE, a framework for learning force-reactive humanoid loco-manipulation.
By Sihang Li · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.