An Efficient and Complete Approach for Cooperative Path-Finding

Cooperative path-finding requires the computation of a set of compatible paths for multiple agents operating on a discrete roadmap. The goal of this problem is to navigate each agent to their unique target vertices without simultaneously occupying the same vertex or edge in the roadmap as any other agent. Such a formulation has applications in warehouse management, transportation networks, (dis)assembly, robotic mining, space exploration, and computer games. The problem of cooperative path-finding has been extensively studied in the literature, with coupled techniques that attempt to intelligently prune the search space into something more tractable while maintaining completeness, as well as decoupled techniques that plan for each agent independently and utilize sophisticated heuristics in order to avoid collisions along these paths. Coupled search techniques have been employed to separate large problems into fully coupled sub-problems (van den Berg et al. 2009), or segment the roadmap into smaller topologies with known characteristics (Ryan 2007) to solve the composite problem sequentially. Decoupled techniques employ heuristics that prioritize agents (Erdmann and Lozano-Perez 1986), or tune the velocities on precomputed paths (Kant and Zucker 1986). Modern approaches consider dynamic prioritization and windowed search (Silver 2005), or domain restriction to guarantee completeness and tractability (Wang and Botea 2009).