Motion-Driven Action-Based Planning

Abstract

Achievement of robotic goals generally needs both plan synthesis and plan execution through physical motions. Costs of actions in robotic tasks are generally motion-dependent. Generally there are many action-based plans for achieving a goal and usually there are many motion plans for executing each action-based plan. Many efficient action-based planners and motion planners have been developed in the last twenty years. One can exploit these computational advances to find low-cost motion plans from the space of motion plans for executing a large number of action-based plans. In this paper we report on generation of action-based plans with low motion-related cost for their execution. We report on empirical evaluation which shows that the motion-related costs for executing action-based plans found with our approach are lower than those for action-based plans found with no motion cost information available to the action-based planner.