Distributed Strategy-making Method in Multiple Mobile Robot System

Abstract

A distributed algorithm for making strategies in multiple autonomous system is presented in this paper. Here, every robot in the system has several tactics to be selected, and a set of probabilities to select each tactic is called a strategy. The problem is “how does each robot get one’s proper strategy independently by trying tactics, getting responses from the environment around itself and revising each strategy?” From the viewpoint of simplicity of modeling and analysis, authors adopt reinforcement-learning-like approach. Each robot memorizes “evaluated pay-off values for every tactic” and the “strategy,” and revises those values asymptotically by iterating four steps as follows: (1) select a certain tactic by reflecting the present strategy of its own and try it, (2) get pay-off value for the applied tactic from the environment, (3) revise the estimated pay-off value of tactics, (4) revise the strategy by using (3). Path-selecting simulation of multiple mobile robots, in which fifty robots move between two areas through two paths, is made to verify effectiveness of the proposed method. By evaluating turnaround time of robots with variety of path widths′ ratios, the proposed algorithm is shown to be superior to other algorithms such as one-way-traffic-strategy. Convergence to global optimal solution derived by the proposed method is discussed for simplified situation by using payoff matrix in theory of game.