Improving generalization in visual reinforcement learning through data distribution

doi:10.1016/j.eswa.2026.132126

Improving generalization in visual reinforcement learning through data distribution

2026-03-19

0

PRE

AI

DOI：10.1016/j.eswa.2026.132126

Original

Original request for help

Save

Abstract

En 中文

• Derived a theoretical upper bound of generalization error in visual RL. • Analyzed how background variation impacts policy generalization. • Proposed two auxiliary losses to improve generalization in visual RL. • Verified effectiveness through extensive experiments on diverse benchmarks.

Keywords:

generalization

visual reinforcement learning

background variation

auxiliary losses

data distribution

Improving generalization in visual reinforcement learning through data distribution

Abstract

Journal

Organization

Cited Papers