Multi-robot motion planning gets 10x faster with new hierarchical decomposition method

Coordinating multiple robots to avoid collisions without bogging down in exponential computation has long been a core challenge in robotics. A team from the University of Illinois Urbana-Champaign (Ngui, McBeth, Motes, Morales, Amato) now presents a solution that slashes planning time by up to an order of magnitude. Their method, detailed in a paper accepted to WAFR 2026, introduces hierarchical subproblem expansion and workspace decomposition refinement. Instead of searching the vast joint configuration space (which grows combinatorially with each robot), the planner first breaks the workspace into discrete regions. It then uses a fast discrete search over those regions to decide when robots need to coordinate and when they can move independently. By iteratively refining the workspace decomposition, the planner can later solve smaller, decoupled configuration-space subproblems, dramatically reducing computation.

Prior work had used workspace topology to guide coordination, but this new approach takes a further step: it dynamically adapts the resolution of the workspace representation during planning. The result is a planner that scales to larger robot teams without the usual explosion in planning time. The authors demonstrate significant speedups in simulation, with planning times often an order of magnitude faster than baseline methods. While the work is theoretical and not yet deployed on physical robots, it points toward a practical path for real-time coordination of robot swarms in warehouses, factories, and autonomous vehicle fleets. The full paper is available on arXiv (2605.20395).