We propose an approach to solve multi-agent path planning (MPP) problems for complex environments. Our method first designs a special pebble graph with a set of feasibility constraints, under which MPP problems have feasibility guarantee. We further propose an algorithm to greedily improve the optimality of planned MPP solutions via parallel pebble motions. As a second step, we develop a mesh optimization algorithm to embed our pebble graph into arbitrarily complex environments. We show that the feasibility constraints of a pebble graph can be converted into differentiable geometric constraints, such that our mesh optimizer can satisfy these constraints via constrained numerical optimization. We have evaluated the effectiveness and efficiency of our method using a set of environments with complex geometries, on which our method achieves an average of 99.0% free-space coverage and 30.3% robot density within hours of computation on a desktop machine.
翻译:我们提出了解决复杂环境多试剂路径规划(MPP)问题的方法。我们的方法是首先设计一个带有一系列可行性限制的特殊石块图,在这种限制下,MPP问题有可行性保证。我们进一步建议一种算法,通过平行的石块运动,贪婪地改进计划中的MPP解决方案的最佳性。作为第二步,我们开发了一种网状优化算法,将我们的石块图嵌入任意复杂的环境中。我们表明,小块图的可行性限制可以转化成不同的几何限制,这样我们的网状优化器就能通过有限的数字优化来满足这些限制。我们用一套具有复杂地理特征的环境评估了我们方法的效能和效率,我们的方法在台式机器上平均在数小时内实现99.0%的自由空间覆盖和30.3%的机器人密度。