.png){kind=icon}
.png){kind=icon}
Return
Enhancing environmental modeling and maximum diffusion reinforcement learning using evolutionary computation for optimal performance
.png){kind=icon}
PRE
AI
DOI：10.1016/j.asoc.2025.114111.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• To address the limitations of environmental model accuracy in model-based reinforcement learning, this paper introduces an evolutionary algorithm integrated into the Maximum Diffusion Reinforcement Learning (MaxDiff RL) framework. The method generates a population of perturbed environment models, enhancing the search for more accurate models and improving overall performance. • The proposed evolutionary approach optimizes environment models by combining gradient-free evolutionary algorithms with gradient-based optimization. This method demonstrates improved performance and sample efficiency across robotic continuous control tasks, outperforming baseline algorithms.