Physics-Informed Policy Optimization via Analytic Dynamics Regularization

TL;DR

PIPER is a physics-informed reinforcement learning framework that incorporates analytical physics constraints into policy optimization, improving sample efficiency and physical consistency in robotic control.

Contribution

The paper introduces PIPER, a novel RL method that seamlessly integrates analytical physics constraints into neural policy training without modifying existing algorithms or simulators.

Findings

Enhanced learning efficiency and stability in robotic control tasks.

Significantly improved control accuracy with physics-informed regularization.

No need for changes to existing simulators or RL algorithms.

Abstract

Reinforcement learning (RL) has achieved strong performance in robotic control; however, state-of-the-art policy learning methods, such as actor-critic methods, still suffer from high sample complexity and often produce physically inconsistent actions. This limitation stems from neural policies implicitly rediscovering complex physics from data alone, despite accurate dynamics models being readily available in simulators. In this paper, we introduce a novel physics-informed RL framework, called PIPER, that seamlessly integrates physical constraints directly into neural policy optimization with analytical soft physics constraints. At the core of our method is the integration of a differentiable Lagrangian residual as a regularization term within the actor's objective. This residual, extracted from a robot's simulator description, subtly biases policy updates towards dynamically consistent solutions. Crucially, this physics integration is realized through an additional loss term during policy optimization, requiring no alterations to existing simulators or core RL algorithms. Extensive experiments demonstrate that our method significantly improves learning efficiency, stability, and control accuracy, establishing a new paradigm for efficient and physically consistent robotic control.