Empowerment-based prediction of swarm performance in collective perception

We apply a recently introduced information-theoretic complexity measure for predicting swarm performance in the Collective Perception scenario. Our approach builds on the principle of Empowerment, which offers a task-independent, universal and generic utility. We evaluated the empowerment measure with respect to the performance of one recent state-of-the-art neural network decision-making controller. The results reveal significant strong correlation between the levels of empowerment and both the accuracy and completion time achieved by a simulated swarm operating in nine distinctive environmental patterns and five swarm communication ranges, used previously to assess swarm performance. These findings suggest the potential of empowerment to analytically predict swarm performance and support the design of more efficient decision-making strategies.