Empirical analysis of PGA-MAP-Elites for Neuroevolution in Uncertain Domains

Quality-Diversity algorithms, among which are the Multi-dimensional Archive of Phenotypic Elites (MAP-Elites), have emerged as powerful alternatives to performance-only optimisation approaches they enable generating collections diverse and high-performing solutions an problem. However, often limited low-dimensional search spaces deterministic environments. The recently introduced Policy Gradient Assisted MAP-Elites (PGA-MAP-Elites) algorithm overcomes this limitation by pairing traditional Genetic operator with a gradient-based inspired deep reinforcement learning. This new guides mutations toward using policy gradients (PG). In work, we propose in-depth study PGA-MAP-Elites. We demonstrate benefits PG on performance reproducibility generated when considering uncertain domains. firstly prove that PGA-MAP-Elites is highly performant in both high-dimensional environments, decorrelating two challenges it tackles. Secondly, show addition outperforming all considered baselines, reproducible approaching found built specifically for applications. Finally, ablation analysis dynamic PG-based variation. variation determinant guarantee but only essential during early stage process, where finds regions space.