Robot Path Planning in Crowd Environments via Dynamic Interaction Modeling and Deep Reinforcement Learning

The unstructured nature and high dynamic of crowded public environments introduce critical challenges for mobile robots, particularly in balancing path planning efficiency, human safety and social comfort. Existing approaches often neglect dynamic interaction modeling and quantitative comfort assessment. This paper introduces a stress index derived from deformation analysis of agents' social spaces to quantify human discomfort during robot navigation. Building on this metric, a novel framework integrating a dynamic interaction model with deep reinforcement learning is proposed. By embedding the stress index into the state space, reward function, and value network architecture, the method enables robots to optimizing collision-free trajectories while autonomously penalize socially disruptive behaviors. Simulation results demonstrate that the algorithm outperforms conventional approaches, achieving fluent navigation and reducing human stress levels comparing with baseline models. The proposed stress-aware mechanism effectively bridges the gap between robotic efficiency and human-centric requirements, providing a framework for socially compliant navigation in dynamic crowds.