Multi-agent path finding (MAPF) in realistic environments requires not only collision-free coordination but also adherence to kinodynamic constraints, such as, for example, limits on acceleration, velocity and turning radius. While considerable progress has been made in the development of MAPF algorithms, the influence of different motion primitives on planning performance under such constraints remains underexplored. In this work, we aim to integrate and systematically evaluate several distinct sets of motion primitives within a kinodynamically constrained multi-agent planning framework. We analyze how the choice of primitive set impacts key plan characteristics, including solution quality, feasibility, and computational efficiency, across a range of standard datasets from the MAPF benchmark suite by MovingAI. Our experimental results demonstrate that careful selection of motion primitives enables a favorable trade-off between computational cost and plan quality. Furthermore, we contextualize our findings by comparing this primitive-based planning paradigm with an alternative approach that enforces kinodynamic constraints via post-processing optimization.
DOI: 10.1007/978-3-032-13612-1_41
Download the proceedings from Springer Nature: https://link.springer.com/content/pdf/10.1007/978-3-032-13612-1.pdf
Zavarzin, N., Yakovlev, K. (2026). Empirical Evaluation of Motion Primitives in Multi-Agent Path Finding with Kinodynamic Constraints // Proceedings of the Ninth International Scientific Conference “Intelligent Information Technologies for Industry” (IITI’25), Volume 2, pp. 467–478.