Heuristic search algorithms with applications to path planning on uneven terrains

Abstract

Planning a desirable path between two given locations is an important problem in many real-world applications. Most of these path planning problems can be perceived as graph search problems at an abstract level. Due to higher computational efficiency, A*-like heuristic search algorithms have been favored over uniform-cost search algorithms to solve such problems. This thesis further explores the power of heuristic search algorithms with applications to mobile robot path planning on uneven terrains. Three heuristic search algorithms are proposed. The first of them, Constraints Satisfying A* (CSA*) algorithm, is capable of solving search problems with multiple constraints. When guided by admissible heuristics, CSA* guarantees to return with an optimal solution to a given constrained search problem if such a solution exists. When guided by consistent heuristics, CSA* is optimally efficient over a class of equally informed admissible constrained search algorithms with respect to both the set of paths expanded and the number of expansion operations. The main objective of the other two algorithms is efficient path replanning for navigating in dynamic and uncertain environments. Rapidly Replanning A* (RRA*) algorithm and Rapidly Replanning Multiobjective A* (RRMA*) algorithm are respectively proposed for replanning of single-objective and multiobjective paths. Both RRA* and RRMA* use their previous search results to speed up their replanning processes. Novel methods are proposed to update an existing search tree/graph when changes in the environment are detected. With the proposed methods, RRA* and RRMA* are able to efficiently replan paths by considerably minimizing the number of cost recalculations.