Navigation in Multi-Agent System with Side Preference Path Optimizer

Abstract

Multi-agent path planning aims to find a set of collision-free global path in the joint configuration space to move each agent from the given start states to the goal states and keep the total cost of the path to a minimum. However, the multi-agent systems in the real-world turn out to be a non-communication and partially observable system where the goal states and the planning strategy of other agents are usually unknown. In order to solve the multi-agent path planning problem in a more general environment, a social-aware navigation method is proposed in this paper. By imitating the most basic human road rules, left/right-handed rules, we add side preference to the global path planning by using a path optimizer so that each agent only needs to consider its own planning while the system is approximately optimal. The proposed path optimizer is tested in benchmark maps, robot-robot and robot-human interaction environment, respectively. With the side preference path optimizer, the success rate of multi-agent navigation with 6 agents increase from 0% to 78% at the cost of only extra 0.1 seconds planning time. The proposed method makes the navigation in human-robot interaction scenes become easier.