SCDP: Learning Humanoid Locomotion from Partial Observations via Mixed-Observation Distillation

2026-03-10

AI Summary

This work addresses the challenge of achieving robust motion control for humanoid robots using only onboard sensors without access to full state observations. The authors propose a diffusion-based policy leveraging hybrid observation distillation, which employs constrained denoising, context distribution alignment, and context-aware attention masking during training. By utilizing full state supervision in simulation, the policy learns to implicitly infer motion states solely from historical sensor inputs, enabling end-to-end control without explicit state estimation. In simulation, the method achieves 99–100% success rates in velocity tracking and 93% accuracy in AMASS motion imitation. Furthermore, it demonstrates real-world deployment on the G1 humanoid robot, achieving stable walking at 50 Hz, thereby validating its consistency and practicality across simulation and physical hardware.

Technology Category

• Intelligent Robots: State Estimation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Distilling humanoid locomotion control from offline datasets into deployable policies remains a challenge, as existing methods rely on privileged full-body states that require complex and often unreliable state estimation. We present Sensor-Conditioned Diffusion Policies (SCDP) that enables humanoid locomotion using only onboard sensors, eliminating the need for explicit state estimation. SCDP decouples sensing from supervision through mixed-observation training: diffusion model conditions on sensor histories while being supervised to predict privileged future state-action trajectories, enforcing the model to infer the motion dynamics under partial observability. We further develop restricted denoising, context distribution alignment, and context-aware attention masking to encourage implicit state estimation within the model and to prevent train-deploy mismatch. We validate SCDP on velocity-commanded locomotion and motion reference tracking tasks. In simulation, SCDP achieves near-perfect success on velocity control (99-100%) and 93% tracking success in AMASS test set, performing comparable to privileged baselines while using only onboard sensors. Finally, we deploy the trained policy on a real G1 humanoid at 50 Hz, demonstrating robust real robot locomotion without external sensing or state estimation.

Problem

• humanoid locomotion
• partial observability
• sensor-based control
• state estimation
• offline imitation learning

Innovation

• Sensor-Conditioned Diffusion Policies
• Mixed-Observation Distillation
• Implicit State Estimation
• Partial Observability
• Humanoid Locomotion