Response probability enhances robustness in decentralized threshold-based robotic swarms

In this paper, we investigate how response probability may be used to improve the robustness of reactive, threshold-based robotic swarms. In swarms where agents have differing thresholds, adding a response probability is expected to distribute task experiences among more agents, which can increase the robustness of the swarm. If the lowest threshold agents for a task become unavailable, distributing task experience among more agents increases the chance that there are other agents in the swarm with experience on the task, which reduces performance decline due to the loss of experienced agents. We begin with a mathematical analysis of such a system and show that, for a given swarm and task demand, we can estimate the response probability values that ensure team formation and meet robustness constraints. We then verify the expected behavior on an agent based model of a foraging problem. Results indicate that response probability may be used to tune the tradeoff between system performance and system robustness.