Multi-agent trajectory planning: A decentralized iterative algorithm based on single-agent dynamic RRT*

This paper addresses trajectory planning in a multi-agent cooperative setting, where
$n$
agents are moving in the same region and need to coordinate so as to maintain a certain pairwise safety distance, while avoiding obstacles. We introduce a decentralized strategy that is based on an iterative (re)plan-compare-assign process. The key features of the proposed strategy are that coordination is obtained via the compare-assign phase in at most
$n$
iterations (including the initialization), and (re)planning is performed by the agents using a single-agent planner, considering the tentative trajectories of the others fixed, and without sharing with them their tracking capabilities and adopted cost criterion. In the proposed implementation, each agent uses a dynamic Rapidly exploring Random Tree star (RRT*) planner that integrates a new prune and graft feature to avoid rebuilding a new tree from its root each time replanning is needed. The resulting Multi-RRT* algorithm is tested in 2D scenarios and shows promising results.