Relaxation Heuristics for Multiagent Planning

Book One

Volume

Proceedings of the International Conference on Automated Planning and Scheduling, 24

Full Technical Papers

Similarly to classical planning, in MA-Strips multiagent planning, heuristics significantly improve efficiency of search-based planners. Heuristics based on solving a relaxation of the original planning problem are intensively studied and well understood. In particular, frequently used is the delete relaxation, where all delete effects of actions are omitted. In this paper, we present a unified view on distribution of delete relaxation heuristics for multiagent planning. Until recently, the most common approach to adaptation of heuristics for multiagent planning was to compute the heuristic estimate using only a projection of the problem for a single agent. In this paper, we place such approach in the context of techniques which allow sharing more information among the agents and thus improve the heuristic estimates. We thoroughly experimentally evaluate properties of our distribution of additive, max and Fast-Forward relaxation heuristics in a planner based on distributed Best-First Search. The best performing distributed relaxation heuristics favorably compares to a state-of-the-art MA-Strips planner in terms of benchmark problem coverage. Finally, we analyze impact of limited agent interactions by means of recursion depth of the heuristic estimates.

10.1609/icaps.v24i1.13642