Work

MV*: A kinodynamic multi-agent path planning algorithm in unstructured environments

Year: 2026
Type: article
Abstract: This paper presents MV*, a centralised kinodynamic multi-agent path planner for car-like vehicles that must coordinate in shared, unstructured environments with static obstacles. The method couples a velocity-aware single-agent lattice search, termed V*, with a priority-based coordination layer. Each single-agent search expands dynamically-feasible motions over position, heading, speed, and time, so that route and speed are decided jointly rather than in sequence, and waiting is represented as an explicit zero-velocity state. The coordination layer searches over acyclic priority orderings and uses branch-and-bound pruning to discard priority configurations that cannot improve the incumbent solution. When a conflict is encountered, the search branches based on alternative priority relations among the conflicting agents, while a scoring mechanism guides exploration toward more promising branches. For each branch, lower-priority agents are replanned around the trajectories of higher-priority agents until a conflict-free priority ordering is obtained. On the MVTP benchmark, the planner achieves substantially shorter runtimes while maintaining solution quality close to that of the reference optimisation solver. The scalability of MV* is further demonstrated on the CL-CBS benchmark, where it coordinates up to 100 vehicles within the prescribed time limit.
Authors: Abdullah Zareh Andaryan, Michael G.H. Bell, Mohsen Ramezani, D. Glenn Geers
Language: English
Open Access status: hybrid
Estimated APC paid: $4,040
OpenAlex ID: W7108337373
FWCI:
Cites: 26
Cited by:
Related to: 10