Optimising Robot Swarm Formations by Using Surrogate Models and Simulations

Optimising a swarm of many robots can be computationally demanding, especially when accurate simulations are required to evaluate the proposed robot configurations. Consequentially, size instances and swarms must limited, reducing number problems that addressed. In this article, we study viability using surrogate models based on Gaussian processes artificial neural networks as predictors robots’ behaviour arranged in formations surrounding central point interest. We have trained tested them terms accuracy execution time five different case studies comprising three, five, ten, fifteen, thirty robots. Then, best performing combined with ARGoS been used obtain optimal configurations for by our hybrid evolutionary algorithm, genetic algorithm local search. Finally, obtained unseen scenarios initial positions robustness stability achieved formations. The exhibited speed increases up 3604 respect simulations. optimisation converged 91% runs stable were 79% testing scenarios.