Benchmarking Quality-Diversity Algorithms on Neuroevolution for Reinforcement Learning

Flageat, Manon; Lim, Bryan; Grillotti, Luca; Allard, Maxime; Smith, Simόn C; Cully, Antoine

Abstract

We present a Quality-Diversity benchmark suite for Deep Neuroevolution in Reinforcement Learning domains for robot control. The suite includes the definition of tasks, environments, behavioral descriptors, and fitness. We specify different benchmarks based on the complexity of both the task and the agent controlled by a deep neural network. The benchmark uses standard Quality-Diversity metrics, including coverage, QD-score, maximum fitness, and an archive profile metric to quantify the relation between coverage and fitness. We also present how to quantify the robustness of the solutions with respect to environmental stochasticity by introducing corrected versions of the same metrics. We believe that our benchmark is a valuable tool for the community to compare and improve their findings. The source code is available online.

Citation

Flageat, M., Lim, B., Grillotti, L., Allard, M., Smith, S. C., & Cully, A. (2022, July). Benchmarking Quality-Diversity Algorithms on Neuroevolution for Reinforcement Learning. Paper presented at Gecco 2022, Boston, Massachusetts

• Presentation Conference Type: Conference Paper (unpublished)
• Conference Name: Gecco 2022
• Start Date: Jul 9, 2022
• End Date: Jul 13, 2022
• Deposit Date: Jul 11, 2023
• DOI: https://doi.org/10.48550/arXiv.2211.02193
• Keywords: Quality-Diversity, Evolutionary algorithms
• Public URL: http://researchrepository.napier.ac.uk/Output/3136250