First Deployable Dynamic-CoM: A Unified Policy and Method-Agnostic Benchmark for Humanoid Single-Leg Balance

2026-08-01

AI Summary

Existing general-purpose policies for humanoid robots struggle to achieve static single-leg balance and often rely on post-hoc recovery mechanisms such as stepping or hopping. This work proposes an end-to-end deployable policy that directly embeds dynamic center-of-mass (dynamic-CoM) observations—reconstructed solely from encoder and IMU data—into the actuation pipeline, coupled with an imbalance-prevention-oriented reward mechanism inspired by human postural control. Key contributions include the first online estimation and real-world deployment of dynamic-CoM without external sensors, the establishment of the first method-agnostic and reproducible single-leg balance sim2sim benchmark, and the introduction of asymmetric FastSAC training with a privileged critic. The proposed FDDC policy achieves successful single-leg balance in 86 out of 90 unseen motions and demonstrates zero-shot transfer to the Unitree G1 physical robot. Ablation studies confirm that dynamic-CoM observations are critical, with performance dropping by 40 percentage points when removed.

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 hardware policy because it requires a base linear velocity no on-board sensor provides; 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 by asymmetric FastSAC with a privileged critic and no distillation, the resulting policy, FDDC (First Deployable Dynamic-CoM), holds clean single-leg balance on 86 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 40 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 its training one, a step toward turning balance from a per-task trick into a capability the field can measure.

Problem

• humanoid balance
• single-leg stance
• center of mass
• capture point
• postural control

Innovation

• dynamic-CoM
• single-leg balance
• humanoid locomotion
• method-agnostic benchmark
• capture point