IMPROVING THE PERFORMANCE OF REINFORCEMENT LEARNING AGENTS USING EXPEPERIENCE-BASED CONTEXT
Published in GÉP 2025/2
https://doi.org/10.70750/GEP.2025.2.2
Farkas Péter,
PhD hallgató
Szőke László
PhD hallgató
Dr. Aradi Szilárd
egyetemi docens
Dr. Gyurkó Zoltán
R&D csoportvezető
ABSTRACT
This paper proposes an experience-based online adaptation framework for reinforcement learning agents, enabling them to adjust to changing conditions by leveraging past state-action transitions, improving their performance in dynamic environments without relying on hard-to-obtain information. The performance of our solution is evaluated through the control problem of a robot model covering a wide dynamic range.
