Learning Forward Looking Adaptation to Dynamic Payloads for Quadruped Locomotion Via Physics-Informed Neural Networks
Oscar Youngquist, Hao Zhang
AI summary
Problem
Existing payload-adaptive locomotion methods react too slowly to sudden mass shifts or lack physical consistency, causing poor generalization and instability when transporting dynamic payloads in unstructured environments.
Approach
FLAP combines a physics-informed neural network that predicts future joint states and payload-induced forces with a composite adaptive control law to generate rapid, anticipatory torque compensations in real time.
Key results
- 94.3% average reduction in locomotion failure rates
- Lower torso and joint tracking errors than reactive and learning baselines
- Successful real-world quadruped deployment with dynamic payloads
- Physically consistent dynamics model generalizing to unseen payloads
Why it matters
Enables quadruped robots to safely and reliably transport dynamic payloads in unstructured real-world environments, advancing autonomous inspection, disaster response, and search-and-rescue applications.
Abstract
Payload-adaptive locomotion is an essential capa- bility for quadruped robots operating in real-world scenarios, particularly when tasked with transporting dynamic payloads. Existing approaches face fundamental limitations: reactive adaptation strategies respond too slowly to sudden payload changes, while learning-based methods often yield physically inconsistent models of robot dynamics that generalize poorly to novel states. To address these key challenges, we introduce Forward-Looking Adaptation to Dynamic Payloads (FLAP), a novel approach that learns to proactively compensate for discrepancies between expected and actual locomotion behavior induced by dynamic payloads. FLAP combines two critical components: (1) a physics-informed neural network (PINN) that predicts anticipated joint states while enforcing physical consistency through dynamics based loss functions, and (2) a composite adaptive control law that rapidly generates anticipa- tory joint torque compensations based on the PINN’s predictions. Through unifying structured dynamics modeling with real-time anticipatory control, our method enables generalizable and phys- ically consistent adaptation to dynamic payloads. Experimental results demonstrate that FLAP achieves robust locomotion under diverse payload conditions on physical quadruped robots in real-world environments. More details available on the project website: https://hcrlab.gitlab.io/project/flap.