DR-MPC: Deep Residual Model Predictive Control for Real-world Social Navigation

2024-10-14

arXiv.org

AI Summary

Safe navigation for service robots in high-density human environments remains challenging due to complex social dynamics and stringent safety requirements. Method: This paper proposes a residual MPC-DRL collaborative framework: model predictive control (MPC) provides safety-aware priors, while deep reinforcement learning (DRL) learns intricate social behaviors via a residual policy; we further introduce, for the first time, an out-of-distribution (OOD) state safety estimation module to enhance initial policy safety and cross-scenario generalization. The method achieves end-to-end socially aware navigation using less than four hours of real-world training data. Results: In simulation, our approach outperforms both conventional and residual DRL baselines. Real-robot experiments demonstrate stable operation across diverse high-density dynamic scenarios, with exceptionally low collision rates—validating its safety, robustness, and engineering practicality.

Technology Category

• Natural Language Processing: Safety and Robustness
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Security and Privacy: Large-scale security measurements
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

How can a robot safely navigate around people with complex motion patterns? Deep Reinforcement Learning (DRL) in simulation holds some promise, but much prior work relies on simulators that fail to capture the nuances of real human motion. Thus, we propose Deep Residual Model Predictive Control (DR-MPC) to enable robots to quickly and safely perform DRL from real-world crowd navigation data. By blending MPC with model-free DRL, DR-MPC overcomes the DRL challenges of large data requirements and unsafe initial behavior. DR-MPC is initialized with MPC-based path tracking, and gradually learns to interact more effectively with humans. To further accelerate learning, a safety component estimates out-of-distribution states to guide the robot away from likely collisions. In simulation, we show that DR-MPC substantially outperforms prior work, including traditional DRL and residual DRL models. Hardware experiments show our approach successfully enables a robot to navigate a variety of crowded situations with few errors using less than 4 hours of training data.

Problem

• Safe robot navigation in crowds
• Complex human motion patterns
• Efficient real-world DRL training

Innovation

• Deep Residual Model Predictive Control
• Blends MPC with model-free DRL
• Safety component for collision avoidance