Plan distance heuristics for task fusion in distributed temporal continuous planning

Abstract

Automating planning for large teams of heterogeneous robots is a growing challenge, as robot capabilities diversify and domain complexities are incorporated. Temporal and continuous features accurately model real-world constraints, but add computational complexity. Distributed planning methods, suc h as the Coalition Formation then Planning framework, allocate tasks to robot teams and plan each task separately to accelerate planning. However, the task decomposition limits cooperation between coalitions allocated to different tasks and results in lower quality plans that require more actions and time to complete. Task Fusion estimates couplings between tasks and fuses coupled coalition-task pairs to improve cooperation and produce higher quality plans. Task Fusion relies on existing heuristics, which were ineffective and often resulted in worse results than the baseline framework. This manuscript introduces new heuristics that outperform the existing methods in two complex heterogeneous multi-robot domains that incorporate temporal and continuous constraints.