Simulated and Real-World Evolution of Predator Robots

Abstract

This paper addresses the problem of designing behavioural strategies for a group of robots with a specific task, capturing another robot. Our proposed approach is to employ a "smart" prey with a pre-programmed strategy based on a novel Gaussian model of danger zones and use an evolutionary algorithm (EA) to optimize the predators’ behavior. The EA is applied in two stages: first in simulation, then in hardware on the real robots. The best evolved robot controllers are then further inspected and compared by their robustness, i.e., performance under different conditions. The results show that our approach is successful, combining simulations, real-world evolution, and robustness analysis it is possible to develop good solutions for the predator-prey problem.