Young Researcher and Innovator Conference Grant

Utilizing Evolution Strategies to Train Transformers in Reinforcement Learning

Matyáš Lorenc, Charles University, Prague, Czechia
Roman Neruda, Czech Academy of Sciences, Prague, Czechia

Genetic and Evolutionary Computation Conference (GECCO 2026)
San José, Costa Rica
13-17 July 2026

Grantee: Matyáš Lorenc, Charles University, Prague, Czechia
Type of publication: Extended abstract
Type of presentation: Poster

Start date: 2026-07-13
End date: 2026-07-17
Awarded: 2026-05-18

Abstract

We explore the capability of evolution strategies to train an agent with a policy based on a transformer architecture in a reinforcement learning setting. We performed experiments using OpenAI’s highly parallelizable evolution strategy to train Decision Transformer in the MuJoCo Humanoid locomotion environment and in the environment of Atari games, testing the ability of this black-box optimization technique to train even such relatively large and complicated models (compared to those previously tested in the literature). The examined evolution strategy proved to be, in general, capable of achieving strong results and managed to produce high-performing agents, showcasing evolution’s ability to tackle the training of even such complex models.