Conclusions
We have examined the use of the RL technique to provide a mobile robot with the capability to operate autonomously in hazardous environments in general, autonomously charging its batteries in particular. We have implemented the robot control architecture in a mobile agent environment, which provides software engineering advantages in comparison to an amalgamation of other distributed computing architectures. The proposed system is more adaptable and robust and allows a mobile robot to dynamically position control or learning to reach the most appropriate location within the environment. Experimental results show the learning characteristics of RL control in five different simulated environments. It has been shown that RL can be used for a mobile robot to learn efficient control policies in a range of environments of varying complexity. A comparison of the RL approach against a non-AI control method shows that RL provides a more efficient and safer method for a mobile robot to return to a recharging station. The RL method can be used to successfully guide the mobile robot back to the recharging station even in environments where the reactive method would have failed to find a path. This justifies the implementation of the AI technique in our wider multiple robot architecture