Flow-based Autonomous Mobile Robot Navigation Considering Social Interaction in Complex Crowd Environments

In recent years, autonomous mobile robots have attracted attention in various public spaces, such as urban environments, festivals, and exhibitions, where they perform tasks like goods delivery, guidance, and information provision. However, existing research has focused on obstacle avoidance and shortest‐path planning, showing limitations in maintaining natural and socially acceptable distances while navigating among densely packed pedestrians. In this study, we propose a reinforcement learning–based navigation method that enables a robot to traverse crowded spaces efficiently by utilizing simplified flow information derived from crowd movement directions and local densities. This method estimates surrounding pedestrian density and movement directions in real time from the robot’s onboard sensor data, incorporating them into the state representation so that the robot can perceive its environment more effectively. During action selection, positive rewards are given for movements consistent with pedestrian flow, while penalties are applied for abrupt speed changes or maneuvers that risk collisions; thus, the robot learns to integrate smoothly into the crowd flow. The main contributions of this research lie in the design of the state representation using crowd‐flowinformation and the formulation of the reward structure, which together guide the robot to learn safer and more socially acceptable navigation behaviors in pedestrian environments. The simulation environment comprises scenarios with pedestrian counts of three, five, and ten, corresponding to different density levels. We evaluate performance using metrics such as goal reaching success rate, average travel time, path smoothness, and average distance to pedestrians to reflect social compliance. Experimental results demonstrate that the proposed method yields statistically significant improvements in goal reaching success and travel efficiency compared to baseline approaches. Even in high‐density environments, the robot interacts harmoniously with pedestrian flow, effectively maintaining socially acceptable distances. Additionally, we design a reward structure that ensures stable learning under varying density conditions and verify the generalizability of the learned policy. This study suggests the feasibility of deploying autonomous mobile robots in smart cities and large‐scale event venues to navigate in harmony with people. In future work, we plan to implement the method on a real robot platform and conduct field experiments to evaluate real time responsiveness.