Let the mobile robots learn to navigate in the crowd (WMX)

• Main Author: Tang, Longchen
• Other Authors: Wang Dan Wei
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics
• Online Access: https://hdl.handle.net/10356/154877
• Institution: Nanyang Technological University

With the development of artificial intelligence, autonomous mobile robots have served humans in many fields, replacing humans to complete complex tasks. Navigation is the basic function of mobile robots. To complete designed tasks, mobile robots should be able to effectively navigate in real-world environment and avoid other obstacles in the human or surrounding environment. The goal of the project is to implement an algorithm that allows the robot to navigate safely and quickly in both static and dynamic environments. This project aims to combine the traditional path planning algorithm dynamic window method (DWA) with deep reinforcement learning to enable the robot to navigate in a dynamic environment. Our method combines the superiority of the dynamic window method (DWA) in the dynamic constraints of the robot, and the superiority of the navigation method based on deep reinforcement learning for the processing of dynamic obstacles, which allows the robot to reach the target quickly and safely while meeting the dynamic constraints. We also use a training method based on curriculum learning to accelerate the convergence speed of the agent in a dynamic environment and ensure the robustness of the robot's navigation in different environments. Aiming at the dynamic environment, we also designed a novel reward function to speed up the robot to reach the goal while staying away from obstacles, and the number of collisions is significantly reduced compared with traditional methods. We evaluated the algorithm in a simulated reality environment, and the results showed that compared with the traditional DWA algorithm, our method has increased the navigation success rate by 40%. Keyword: Traditional path planning, deep reinforcement learning, curriculum learning, dynamic environment, obstacle avoidance