Mobile Service Robot Path Planning Using Deep Reinforcement Learning

A mobile service robot operates in a constantly changing environment with other robots and humans. The service environment is usually vast and unknown, and the robot is expected to operate continuously for a long period. The environment can be dynamic, leading to the generation of new routes or the permanent blocking of old routes. The traditional path planner that relies on static maps will not suffice for a dynamic environment. This work is focused on developing a reinforcement learning-based path planner for a dynamic environment. The proposed system uses the deep Q-Learning algorithm to learn the initial paths using a topological map of the environment. In an environmental change, the proposed $\pmb \beta $ -decay transfer learning algorithm trains the agent in the new environment. This algorithm uses experience vs. exploration vs. exploitation-based training depending on the similarity of the old and new environments. The system is implemented on the Robotic Operating System framework and tested using Turtlebot3 mobile robot in the Gazebo simulator. The experimental results show that the reinforcement learning system learns all the routes based on the initial topological map of different service environments with an accuracy of over 98%. A comparative analysis of the $\pmb \beta $ -decay transfer learning and non-transfer learning agents is performed based on various evaluation metrics. The transfer learning agent converges twice faster than the non-transfer learning agent.