Subdimensional Expansion for Multi-Objective Multi-Agent Path Finding

Conventional multi-agent path planners typically determine a that optimizes single objective, such as length. Many applications, however, may require multiple objectives, say time-to-completion and fuel use, to be simultaneously optimized in the planning process. Often, these criteria not readily compared sometimes lie competition with each other. Simply applying standard multi-objective search algorithms finding prove inefficient because size of space possible solutions, i.e., Pareto-optimal set, can grow exponentially number agents (the dimension space). This paper presents an approach bypasses this so-called curse dimensionality by leveraging our prior work framework called subdimensional expansion. One example expansion, when applied A*, is M* was limited objective function. We combine principles dominance expansion create new algorithm named (MOM*), which dynamically couples for only those have "interact" MOM* computes complete set efficiently naturally trades off sub-optimal approximations computational efficiency. Our able find problem instances hundreds solutions A* could within bounded time.