Optimal path planning based on a multi-tree T-RRT* approach for robotic task planning in continuous cost spaces

This paper presents an integrated approach to robotic task planning in continuous cost spaces. This consists of a low-level path planning phase and a high-level Planning Domain Definition Language (PDDL)-based task planning phase. The path planner is based on a multi-tree implementation of the optimal Transition-based Rapidly-exploring Random Tree (T-RRT*) that searches the environment for paths between all pairs of configuration waypoints. A method for shortcutting paths based on cost function is also presented. The resulting minimized path costs are then passed to a PDDL planner to solve the high-level task planning problem while optimizing the overall cost of the solution plan. This approach is demonstrated on two scenarios consisting of different cost functions: obstacle clearance in a cluttered environment and elevation in a mountain environment. Preliminary results suggest that significant improvements to path quality can be achieved without significant increase to computation time when compared with a T-RRT-based implementation