Autonomous navigation for mobile robots using deep reinforcement learning

• Main Author: Gan, Wei Han
• Other Authors: Xie Lihua
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/148913
• Institution: Nanyang Technological University

The emergence of machine learning and artificial intelligence has propelled humankind into a new age of information. In this era, virtually any task can be made easier, optimised, or fully automated by machines, with the implementation of Autonomous Vehicles (AV) gaining rise in popularity. This study aims at ascertaining and executing a deep reinforcement algorithm, namely Proximal Policy Optimisation (PPO) into a multi-robot decentralised framework, directly mapping raw sensor data to an agent’s actions and intents. Contrary to its centralised counterparts, this framework does not require a central server, communication protocols, nor comprehensive information of each agent’s state and intentions. Using Ubuntu 16.04, Stage simulator, and Robot Operating System (ROS) as a general platform, these robots strive to achieve competence in autonomous navigation and collision avoidance. Under a framework of curriculum learning, the policy of the robots is continually refined in a four, eight, and random environment sequentially. With ample simulation time, the robots are able to find an optimal policy in each Stage environment to achieve autonomous navigation, along with balancing time-efficiency and reward gain to achieve collision-free routes. Video simulation can be found at https://www.youtube.com/watch?v=mu2QeJ-lKiY.