Planning collision-free paths for multiple robots traversing a shared space is a problem that grows combinatorially with the number of robots. The naive centralised approach soon becomes intractable for even a moderate number of robots. Decentralised approaches, such as priority planning, are much faster but lack completeness. Previously I have demonstrated that the search can be significantly reduced by adding a level of abstraction [1]. I first partition the map into subgraphs of particular known structures, such as cliques, halls and rings, and then build abstract plans which describe the transitions of robots between the subgraphs. These plans are constrained by the structural properties of the subgraphs used. When an abstract plan is found, it can easy be resolved into a complete concrete plan without further search. In this paper, I show how this method of planning can be implemented as a constraint satisfaction problem (CSP). Constraint propagation and intelligent search ordering further reduces the size of the search problem and allows us to solve large problems significantly more quickly, as I demonstrate this in a realistic planning problem based on a map of the Patrick Port Brisbane yard. This implementation also opens up opportunities for the application of a number of other search reduction and optimisation techniques, as I will discuss. © 2008 Springer Berlin Heidelberg.
CITATION STYLE
Ryan, M. (2008). Constraint-based multi-agent path planning. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 5360 LNAI, pp. 116–127). https://doi.org/10.1007/978-3-540-89378-3_12
Mendeley helps you to discover research relevant for your work.