Authors: Seongwon Lee, James Motes, Isaac Ngui, Marco Morales, Nancy M. Amato

Abstract:
We introduce Lazy-DaSH, a variant of the multi-robot task and motion planning method DaSH, which scales to more than double the number of robots and objects compared to the original method and achieves three orders of magnitude faster planning time when applied to the multi-manipulator object rearrangement problem. We achieve this improvement by lazily validating motions proposed by a new task planning layer while DaSH instead computes motion feasibility of all possible actions, most of which are irrelevant to the task. Our proposed method uses a high-level task query phase to identify the necessary planning spaces for task completion and lazily computes motion feasibility only within these spaces. This approach creates a hierarchical structure between the task planning layer and motion planning layer, connected by a constraint feedback mechanism that informs the feasibility of both tasks and motions. Our strategy reduces both the time required to construct state space representations and the query time by keeping the representation smaller. We demonstrate our method in four different scenarios to highlight the scalability of an increasing number of robots and objects and the constraint feedback mechanism for conflict adaptations. 

Unpublished