Using scouts to predict swarm success rate

Abstract

The scenario addressed here is that of a swarm of agents (simulated robots) that needs to travel from an initial location to a goal location, while avoiding obstacles. Before deploying the entire swarm, it would be advantageous to have a certain level of confidence that a desired percentage of the swarm will be likely to succeed in getting to the goal. The approach taken in this paper is to use a small group of expendable robot ldquoscoutsrdquo to predict the success probability for the swarm. Two approaches to solving this problem are presented and compared - the standard Bernoulli trials formula, and a new Bayesian formula. Extensive experimental results are summarized that measure and compare the mean-squared error of the predictions with respect to ground truth, under a wide variety of circumstances. Experimental conclusions include the utility of a uniform prior for the Bayesian formula in knowledge-lean situations, and the accuracy and robustness of the Bayesian approach. The paper also reports an intriguing result, namely, that both formulas usually predict better in the presence of inter-agent forces than when their independence assumptions hold.