A Change of Frame Makes Balance Observable: Distillation-Free Humanoid Single-Leg Stance

Abstract

Unified humanoid policies handle agile whole-body motion, yet stumble on a simple demand: staying balanced on one leg. On our single-leg-balance benchmark, eight released state-of-the-art general policies hold a clean single-leg stance on 0 of 90 test motions; they stay up only by stepping or hopping, recovering from imbalance rather than preventing it. Prevention needs the capture point (xCoM), the center of mass (CoM) extrapolated by its velocity, which has never driven a learned hardware policy because it requires a base linear velocity that no on-board sensor measures directly. A change of frame makes it observable: expressed relative to the support foot, that velocity cancels exactly, leaving an observation reconstructible from encoders and IMU alone. We put this first deployable dynamic-CoM observation directly into the actor that runs on hardware, and pair it with a reward library translated term by term from human postural control, under one principle: prevention over repair. Trained via asymmetric FastSAC without distillation, the resulting policy, DDC (Deployable Dynamic-CoM), holds clean single-leg balance on 89 of 90 held-out motions across nine stratified pose classes and transfers to a real Unitree G1; in ablation, the dynamic-CoM observation is the single largest driver: removing it alone costs 43 points of clean single-leg balance. We release the full stack with the first method-agnostic, reproducible sim2sim benchmark for humanoid single-leg balance, scoring each policy in a simulator distinct from the one it was trained in, to help turn balance from a per-task trick into a capability the field can measure and build in.