Hybrid approaches for mobile robot navigation

The work described in this thesis contributes to the efficient solution of mobile robot navigation problems. A series of new evolutionary approaches is presented. Two novel evolutionary planners have been developed that reduce the computational overhead in generating plans of mobile robot movements. In comparison with the best-performing evolutionary scheme reported in the literature, the first of the planners significantly reduces the plan calculation time in static environments. The second planner was able to generate avoidance strategies in response to unexpected events arising from the presence of moving obstacles. To overcome limitations in responsiveness and the unrealistic assumptions regarding a priori knowledge that are inherent in planner-based and a vigation systems, subsequent work concentrated on hybrid approaches. These included a reactive component to identify rapidly and autonomously environmental features that were represented by a small number of critical waypoints. Not only is memory usage dramatically reduced by such a simplified representation, but also the calculation time to determine new plans is significantly reduced. Further significant enhancements of this work were firstly, dynamic avoidance to limit the likelihood of potential collisions with moving obstacles and secondly, exploration to identify statistically the dynamic characteristics of the environment. Finally, by retaining more extensive environmental knowledge gained during previous navigation activities, the capability of the hybrid navigation system was enhanced to allow planning to be performed for any start point and goal point.