Mission Planning for Field Robots using Symbolic Planning and Topology

Field robotics is an area of research that takes the discipline of robotics from the confines of the laboratory into the unstructured and complex environment of the real world. Planning and guidance systems have been developed to allow field robotic platforms to operate in unstructured environments, but the limited amount of computing resources has constrained the ability of field platforms to dynamically replan their missions. Domain specific planning systems for path planning provide the efficiency that is required to handle large and complex environments, but deliberative higher level mission planning systems typically use a domain independent planner to find a solution to the vehicle’s task. As such, mission planners lack understanding of their spatial environment. This thesis chronicles the development of a belief compression method using topological thinning to simplify the spatial environment sufficiently for it to be solved by a domain independent planner allowing a vehicle’s mission to be planned using information about its spatial environment. Algorithms are evaluated using both simulated and real-world data showing that topological thinning can produce compact domains while maintaining a high level of routing efficiency, enabling the solution of the high-level mission planning problem. This thesis also examines the properties of topological belief compression and the effectiveness of path planning with non-uniform action costs using domain independent planners. To demonstrate the effectiveness of these algorithms, a planning and guidance system is tested on an Autonomous Surface Vessel (ASV) built around a five-metre Wave Adaptive Modular Vehicle platform (WAM-V). When performing simulated rescue tasks for 20 survivors before returning to a dock, the Symbolic With Re-finement planner demonstrated plan generation resulting in a mean reduction in path length of approximately 15% when compared to a Greedy planning system.