Centralized and Distributed Path Planning for Multi-Agent Exploration

Abstract

In this paper an innovative planner for multi-agent exploration problems is presented. The problem to be solved is is closely related to a Multiple Traveling Salesman problem: we seek optimal vehicle routes (shortest paths) for visiting n given targets with m vehicles. The proposed method first considers path planning for a single agent, solving the so-called Subtour problem, where the vehicle should visit k out of the n targets with the shortest possible path. A Genetic Algorithm is implemented to find near-optimal solutions of this Subtour problem. These solutions are then used to create an initial multi-vehicle plan, that is further optimized by a novel evolutionary algorithm to create a good overall team strategy. Results are presented to demonstrate the success of the approach.