AUTHOR=Pagliuca Paolo , Milano Nicola , Nolfi Stefano 

TITLE=Efficacy of Modern Neuro-Evolutionary Strategies for Continuous Control Optimization

JOURNAL=Frontiers in Robotics and AI

VOLUME=Volume 7 - 2020

YEAR=2020

URL=https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2020.00098

DOI=10.3389/frobt.2020.00098

ISSN=2296-9144

ABSTRACT=We analyze the efficacy of modern neuro-evolutionary strategies for continuous control optimization. Overall the results collected on a wide variety of qualitatively different benchmark problems indicate that these methods are generally effective and scale well with respect to the number of parameters and the complexity of the problem and are relatively robust with respect to the setting of hyper-parameters. The comparison of the most promising methods indicates that the OpenAI-ES algorithm outperforms or equals the other algorithms on all considered problems. Moreover, we demonstrate how the reward functions optimized for reinforcement learning methods are not necessarily effective for evolutionary strategies and vice versa. This finding has important consequences since it implies that the comparison performed to date are biased toward one or the other class of algorithm and since it might lead to reconsideration of the relatively efficacy of the two classes of algorithms