FLORE: Learning Force-Reactive Humanoid Loco-Manipulation
Field: Machine Learning Authors: Sihang Li, Xiaofeng Guo, Jinkun Cao, Charlie Cheng, Ruihan Yang et al. Published: 2026-10-26 arXiv: 2610.35862
Abstract
Many 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. We collect robot-free demonstrations using two force-sensing UMI-style handheld grippers for hand motion and contact wrenches, and wearable trackers for body motion. A force-reactive diffusion policy takes vision, proprioception, and measured wrenches as input, then predicts actions and per-hand force commands. A hybrid compliance whole-body controller with force tracking executes these commands while maintaining balance: an integral of the force error measured by real force/torque (F/T) sensors lets each hand track force accurately along the commanded direction and remain compliant along the others. In simulation, the controller tracks per-hand force commands of varied magnitude and direction, with little loss in whole-body motion accuracy. Across three contact-rich tasks on the robot, the force-reactive policy succeeds in 80% of rollouts, compared with 26.7% for a position-only policy trained on the same demonstrations and executed by the same controller.
Key Components
- Robot-free data collection: Two force-sensing UMI-style handheld grippers capture hand motion and contact wrenches; wearable trackers capture body motion — no robot required during demonstrations.
- Force-reactive diffusion policy: Takes vision, proprioception, and measured wrenches as input; predicts actions plus per-hand force commands.
- Hybrid compliance whole-body controller: Maintains balance while tracking forces via an integral of force error measured by real F/T sensors — accurate force tracking along the commanded direction, compliance along all others.
- Simulation: Tracks per-hand force commands of varied magnitude and direction with little loss in whole-body motion accuracy.
- Real robot: 80% success rate across three contact-rich tasks, versus 26.7% for a position-only policy trained on the same demonstrations and executed by the same controller.