Reinforcement learning based path planning using a topological map for mobile service robot

A mobile service robot is a class of robots that operates in a dynamic environment with other robots and human beings. The service environment is often considered large and unknown, and the robot is expected to operate lifelong. The static path planning system that directly uses maps or explores an unknown environment will work fine if the environment does not change over time. But for a service environment, there can be changes in the environment over time, i.e., new routes can be generated, or old routes are permanently blocked. Considering these facts, it is evident that the path planning system for a service robot should be an artificial intelligence-based system that will be able to plan the path given less or no information about the environment and cope with the environment’s changes for a lifelong. Our work aims at developing a path planning framework using reinforcement learning for a service environment. The proposed path planning algorithm uses a Q-Learning algorithm to learn the paths using a topological map of the environment. The system is trained and tested for two different service environments. The experiment results show that the system can learn and generate efficient routes based on the topological maps with an accuracy of 95%. This artificial intelligence-based framework can extend the path planning system’s capability to adapt to long-term route changes.