Layered task planning for mobile manipulators

Abstract

Task planning for mobile manipulators is a challenging problem in robotics. Mobile ma- nipulators can perform a wide range of tasks in complex environments. They must process the symbolic knowledge in order to find a sequence of actions that achieves the specified goals. This paper proposes a hybrid approach named Planning with Shortcuts. It uses a data structure called Layered Task Graph (LTG) which groups primitive actions, robot skills and task patterns in a strictly hierarchical manner. It then uses the LTG to create shortcuts in the state space in order to accelerate the search. A task planner is called to find a set of actions that achieves the goal task. The result of the search is a sequence of high level actions and primitive actions. The high level actions are then decomposed using the task decomposition approach.