Geometry-based distributed arc-consistency method for multiagent planning and scheduling

This research focuses on building a distributed algorithm for planning and scheduling multiple agents to help people deal with events beyond their cognitive capacity, such as car assembly, factory management, spacecraft constellation, etc. We address not only the efficiency of the algorithm but also communication and the individual privacy. As to reason over the problems with multiple agents which are distributed but interconnected, a formal account of the Action-centric Multiagent Simple Temporal Problem (AMSTP) is put forward using the representation of geometries. The key technique we build on is a novel distributed arc-consistency algorithm centered by the geometric method called GDAC, which pays attention to how an agent’s local subproblem affects other agents’ subproblems. The GDAC is based on geometries taking the action rather than the timepoint as a variable, which can deal with continuous intervals and decrease the number of variables. Comprehensive experiments are run and the proposed technique outperforms the competitor and shows considerable merit compared to the centralized algorithm.