Evolutionary approaches to mobile robot systems

This thesis presents an investigation into the applicability of evolutionary computing techniques to problems in the robotics domain. Particular attention is given to the techniques of genetic algorithms (GA), applied to mobile robot path planning and genetic programming (GP) applied to the use of communication within controllers of collaborative teams of mobile robots. The thesis identifies and demonstrates in greater depth some of the key issues affecting the development of robust evolutionary based path-planning systems and demonstrates possible ways for maximising performance resulting from these issues. It also verifies, and in some cases cast doubt on existing results and commonly held beliefs with respect to the evolution of communicating controllers, as well as extending the scope of work in the area of evolution of controllers in communication based environments. The thesis illustrates the GP as a capable tool in this area and in the general area of control system extension and fault-tolerance system development. This work shows that the use of evolutionary based approaches in the development of robotic systems is a viable alternative to existing methods, offering as its strong points three key features. Firstly, the ability to produce a diversity of potential solutions for a task, some offering general characteristics while others specific. Secondly, the ability to determine decompositionallevels of a task both at a functional and a communication control level. Thirdly, the ability to produce optimal or near optimal solutions appropriate to the circumstance or information content. The key results from the development of the GA based path planner show that representation dictates the effectiveness of path generation, with flexibility in its structure being the main pre-requisite. It also shows that issues of robustness can be tackled through the application of methods, which apply a control mechanism allowing for the expansion and contraction of the length of paths when appropriate, or methods that offer improved environment pre-processing. The application of the GP to the evolution of communication based controllers for teams of mobile robots, shows that the evolutionary process can effectively manipulate low and high level functional units. Further, it shows that it is possible for the evolutionary process to identify the most appropriate information content for a task and how best to use it. It also showed that communication can be used in various ways (overriding, questioning and controlling) and that appropriate communication topologies as well as rates of communication can be established. As wells as the aforementioned results an additional finding was the fact that the information contained within the communication did not necessarily have to be task specific, in order for the evolutionary process to make use of it (althou$h this in some cases increased the rate at which it is incorporated). ~