Dependable policy improvement for intelligent agents in new environments

Yao Li and Zhenglin Liang

Reliability Engineering and System Safety, 2025, vol. 261, issue C

Abstract: Intelligent agents often encounter challenges in balancing safety and performance when transitioning from general training scenarios to specific task scenarios due to unknown environmental differences. Under the uncertainty of new scenarios, safety considerations constrain extensive exploration, resulting in limited policy improvement. This paper proposes a novel reinforcement learning approach featuring a dependable policy improvement algorithm that emphasizes safety and confidence throughout the entire training process. The proposed algorithm enhances the baseline policy developed in general training scenarios to guide exploration and designs confidence bounds to evaluate both task performance and safety. By cautiously exploring and updating policies based on data confidence bounds, the approach ensures reliable agent behavior in new, uncertain, and potentially risky environments. Simulation experiments with an automatic guided vehicle (AGV) demonstrate the effectiveness of this approach across various scenarios.

Keywords: Reinforcement learning; Uncertain environment; Safe exploration; Confidence bound; Dependable policy improvement (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0951832025003175
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:reensy:v:261:y:2025:i:c:s0951832025003175

DOI: 10.1016/j.ress.2025.111116

Access Statistics for this article

Reliability Engineering and System Safety is currently edited by Carlos Guedes Soares

More articles in Reliability Engineering and System Safety from Elsevier
Bibliographic data for series maintained by Catherine Liu ().