PvP: Data-Efficient Humanoid Robot Learning with Proprioceptive-Privileged Contrastive Representations

2025-12-15

AI Summary

To address the low sample efficiency and partial observability challenges in whole-body control (WBC) for humanoid robots, this paper proposes a proprioception–privileged-state complementary contrastive representation learning framework. Our method jointly models proprioceptive signals and privileged states—without manual data augmentation—to learn compact, task-relevant latent representations. We introduce the novel paradigm of “proprioception-privileged contrastive learning” and present SRL4Humanoid, the first modular, unified evaluation framework for state representation learning tailored to humanoid robotics. Experiments on the LimX Oli platform demonstrate that our approach accelerates convergence by 2.3× and improves task success rates by 37% in velocity tracking and motion imitation tasks, while significantly enhancing robustness and sample efficiency under dynamic environmental conditions.

Technology Category

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

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Achieving efficient and robust whole-body control (WBC) is essential for enabling humanoid robots to perform complex tasks in dynamic environments. Despite the success of reinforcement learning (RL) in this domain, its sample inefficiency remains a significant challenge due to the intricate dynamics and partial observability of humanoid robots. To address this limitation, we propose PvP, a Proprioceptive-Privileged contrastive learning framework that leverages the intrinsic complementarity between proprioceptive and privileged states. PvP learns compact and task-relevant latent representations without requiring hand-crafted data augmentations, enabling faster and more stable policy learning. To support systematic evaluation, we develop SRL4Humanoid, the first unified and modular framework that provides high-quality implementations of representative state representation learning (SRL) methods for humanoid robot learning. Extensive experiments on the LimX Oli robot across velocity tracking and motion imitation tasks demonstrate that PvP significantly improves sample efficiency and final performance compared to baseline SRL methods. Our study further provides practical insights into integrating SRL with RL for humanoid WBC, offering valuable guidance for data-efficient humanoid robot learning.

Problem

• Improves sample efficiency in humanoid robot reinforcement learning
• Learns compact task-relevant representations without manual data augmentation
• Enables faster and more stable whole-body control policy learning

Innovation

• Proprioceptive-Privileged contrastive learning for compact representations
• No hand-crafted data augmentations for stable policy learning
• Unified modular framework for state representation learning evaluation