Decentralized Structural-RNN for Robot Crowd Navigation with Deep Reinforcement Learning

2020-11-09

IEEE International Conference on Robotics and Automation

AI Summary

Safe and efficient autonomous navigation for mobile robots in dense, partially observable crowd environments remains challenging due to dynamic, heterogeneous agent interactions and limited observability. Method: This paper proposes a Decentralized Structured Recurrent Neural Network (DS-RNN), integrating graph-based relational modeling with structured temporal modeling to enable end-to-end decision-making without prior dynamical assumptions or expert supervision. Contribution/Results: DS-RNN is the first to combine structured RNNs with a decentralized architecture, enabling zero-shot sim-to-real transfer. Trained within the PPO framework, it significantly outperforms state-of-the-art methods in high-density, dynamic crowd scenarios in simulation and generalizes successfully to real-world deployment on a TurtleBot 2i platform—demonstrating strong robustness, scalability, and practical applicability.

Abstract

Safe and efficient navigation through human crowds is an essential capability for mobile robots. Previous work on robot crowd navigation assumes that the dynamics of all agents are known and well-defined. In addition, the performance of previous methods deteriorates in partially observable environments and environments with dense crowds. To tackle these problems, we propose decentralized structural-Recurrent Neural Network (DS-RNN), a novel network that reasons about spatial and temporal relationships for robot decision making in crowd navigation. We train our network with model-free deep reinforcement learning without any expert supervision. We demonstrate that our model outperforms previous methods in challenging crowd navigation scenarios. We successfully transfer the policy learned in the simulator to a real-world TurtleBot 2i.

Problem

• Robot Navigation
• Crowd Environment
• Autonomous Systems

Innovation

• DS-RNN
• Autonomous Learning
• Virtual to Real World Transfer