Persistent Execution of Continuous Time Multi-Robot Pathfinding

Many one-shot Multi-Robot Path Finding (MRPF) algorithms have been able to find collision-free paths for hundreds or even thousands of robots moving to their respective goal positions in dense scenarios, and some persistent execution frameworks are able to drive MRPF persistently. However, these frameworks have great limitations in practical applications, which are based on discrete time and grids. In this paper, we present a persistent execution framework independent of MRPF solvers, which can not only operate in discrete time and grids, but also solve the MRPF problem based on roadmap in continuous time. Our presented prediction replanning method is at the core of the framework driving the operation of the MRPF algorithm in the case of persistent tasks, and it also deals with the impact of algorithm computation time and message transmission delay of the multi-robot system. We also presented a dynamic map addition method to solve the problem that replanning cannot be performed when robot is not on map vertex during the runtime. We evaluate our framework using Continuous-time conflict-based search (CCBS) as the MRPF solver and the results show that it can efficiently drive the MRPF algorithm to run under continuous-time persistent task situations.