EPO: Policy Orchestration at Control Rate

Loading video
Loading videoEPO learns a differentiable energy for every frozen whole-body policy, then selects and refines the most feasible policy-goal pair at control rate, improving precision-robustness with far fewer switches.
HumanoidWhole-Body ControlPolicy OrchestrationRobot LearningEnergy-Based ModelsUnitree G1CoRL 2026MIT
Category: humanoid
Author: @Weirui_Ye
Date: 2026-11-08T00:00:00
Duration: 51.1s
Reference: https://arxiv.org/abs/2611.02546





