Robot-Assisted Pedestrian Regulation Based on Deep Reinforcement Learning.

Area of Science:

• Robotics
• Artificial Intelligence
• Crowd Dynamics

Background:

• Pedestrian regulation is crucial for preventing accidents and enhancing safety in crowded environments.
• Mobile robots are increasingly used for crowd management, leveraging passive human-robot interaction (HRI) to guide pedestrian flow.
• Optimizing pedestrian flow through bottlenecks presents a significant challenge due to complex human motion dynamics.

Purpose of the Study:

• To formulate a robot motion planning problem for optimizing two merging pedestrian flows at a bottleneck exit.
• To develop a deep neural network model for mapping environmental perception to robot motion decisions.
• To train the robot motion planner using deep reinforcement learning for adaptive control.

Main Methods:

• Formulation of a robot motion planning problem for merging pedestrian flows.
• Implementation of a deep neural network to model human-robot interaction dynamics.
• End-to-end training of the motion planner with a deep reinforcement learning algorithm.
• Validation through simulated experiments to evaluate performance.

Main Results:

• The robot successfully identified optimal motion strategies to maximize pedestrian outflow.
• Significant increases in accumulated pedestrian outflow were observed compared to unmanaged or randomly managed flows.
• The deep reinforcement learning approach effectively handled complex dynamic environments without manual feature engineering.

Conclusions:

• The proposed deep reinforcement learning-based robot motion planner is effective for optimizing pedestrian flow in bottleneck scenarios.
• This approach enhances crowd safety and efficiency by maximizing throughput.
• The method demonstrates the potential of AI in managing complex crowd dynamics through passive HRI.