Dynamics Aware Quadrupedal Locomotion via Intrinsic Dynamics Head

2026-05-01

AI Summary

This work addresses the challenge of achieving efficient and stable locomotion in quadrupedal robots over complex terrain by effectively leveraging underlying dynamics. The authors propose a novel training framework that jointly optimizes a control policy and an Intrinsic Dynamics (ID) head in simulation. The ID head learns a mapping from system states to joint torques and guides the policy via a dynamics-aware reward to produce smoother, more predictable behaviors. By incorporating a tunable coefficient, the ID head enables the policy to explicitly reason about physical dynamics, thereby refining its actions and improving convergence quality. Integrating reinforcement learning, dynamics modeling, and sim-to-real transfer, the approach demonstrates significant real-world performance gains: 16.8% higher torque efficiency, 18.6% increased motion frequency, 12.8% greater mechanical power output, and a 6.4% improvement in safe torque utilization.

Abstract

Quadrupedal locomotion plays a critical role in enabling agile, versatile movement across complex terrains. Understanding and estimating the underlying physical dynamics are essential for achieving efficient and stable quadrupedal locomotion. We propose a novel training framework for quadrupedal locomotion that enables the Control Policy to understand and reason about physical dynamics. In simulation, we concurrently train an Intrinsic Dynamics (ID) Head that learns state-to-torque dynamics alongside the Control Policy, and we define a dynamics reward enabled by the ID Head that encourages the Policy toward more predictable dynamical behavior. We also provide a mechanism to tune the learned dynamics in the resulting Policy by controlling the training coefficients of the ID Head. Our simulation experiments show that this mechanism drives convergence to better optima across a wide range of standard quadrupedal locomotion rewards, yielding more efficient and smoother policies. Our real-robot experiments demonstrate sim-to-real transfer of these improvements, with significant gains in torque efficiency (16.8%), action rate (18.6%), and mechanical power (12.8%), while improving safe torque occupancy by 6.4%.

Problem

• quadrupedal locomotion
• physical dynamics
• dynamics estimation
• efficient movement
• stable locomotion

Innovation

• Intrinsic Dynamics Head
• Quadrupedal Locomotion
• Dynamics-aware Control
• Sim-to-Real Transfer
• Reinforcement Learning