Physics-informed Neural Network Predictive Control for Quadruped Locomotion

2025-03-10

AI Summary

High-precision motion control of quadruped robots under unknown payloads remains challenging due to model uncertainty and real-time computational constraints. Method: This paper proposes an online load identification–driven physics-informed neural network (PINN) predictive control framework. It integrates online mass parameter estimation directly into the PINN loss function and employs the dynamics-constrained PINN as an efficient surrogate model for nonlinear model predictive control (NMPC). The approach synergistically combines quadruped dynamics modeling, real-time parameter estimation, and PINN-NMPC co-optimization. Contribution/Results: To the best of our knowledge, this is the first work embedding online mass identification into the PINN loss while ensuring physical consistency. Experimental results demonstrate a 35% improvement in pose tracking accuracy under dynamic payloads of 25–100 kg, significantly faster convergence than state-of-the-art adaptive methods, and millisecond-level real-time optimization capability—fully satisfying stringent real-time control requirements.

Technology Category

• Intelligent Robots: State Estimation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling

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
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

This study introduces a unified control framework that addresses the challenge of precise quadruped locomotion with unknown payloads, named as online payload identification-based physics-informed neural network predictive control (OPI-PINNPC). By integrating online payload identification with physics-informed neural networks (PINNs), our approach embeds identified mass parameters directly into the neural network's loss function, ensuring physical consistency while adapting to changing load conditions. The physics-constrained neural representation serves as an efficient surrogate model within our nonlinear model predictive controller, enabling real-time optimization despite the complex dynamics of legged locomotion. Experimental validation on our quadruped robot platform demonstrates 35% improvement in position and orientation tracking accuracy across diverse payload conditions (25-100 kg), with substantially faster convergence compared to previous adaptive control methods. Our framework provides a adaptive solution for maintaining locomotion performance under variable payload conditions without sacrificing computational efficiency.

Problem

• Precise quadruped locomotion with unknown payloads
• Real-time optimization for complex legged dynamics
• Adaptive control under variable payload conditions

Innovation

• Integrates online payload identification with PINNs
• Embeds mass parameters in neural network loss
• Uses physics-constrained neural surrogate model