גיא שני

In the multi-agent path finding (MAPF) problem, a group of agents search in a graph for a path for each agent where no two paths collide. This work considers MAPF applications in which the agents do not wish to share their paths due to privacy constraints. We formulate two types of privacy constraints in this context: planning-level privacy and execution-level privacy. The former means the agents cannot identify the planned location of the other agents, and the latter means agents cannot sense the location of each other during execution. We show a general approach to preserve planning-level privacy and show to how adapt two popular MAPF algorithms, namely PIBT and LaCAM, to preserve execution-level privacy. We also propose a post-processing technique that allows agents to reduce the costs of the returned solution without losing any privacy.