Confidence-Aware Agentic RL with Failure Risk Uncertainty Bounds for Web Automation
International Journal of Information and Communication Technology Trends · 25.03.2026 · 6 (1) · 28-34
DOI: 10.66525/ijictt70 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Information and Communication Technology Trends, online erschienen am 25.03.2026. Verfasst von Sebastian Schmidt und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Failure risk estimates in web automation are often noisy due to limited observations and rare catastrophic events. This study introduces a confidence-aware agentic reinforcement learning framestudy that explicitly accounts for uncertainty in failure risk estimation. Instead of optimizing point estimates, the agent reasons over confidence intervals of failure probability and cost consumption. Policy updates penalize decisions whose upper confidence bounds exceed acceptable safety thresholds. The framestudy is evaluated on 1,200 web tasks with induced uncertainty through partial observability and stochastic UI behavior. Quantitative results demonstrate that confidence-aware decision-making reduces unexpected failures under deployment-time uncertainty while maintaining stable success rates and cost budgets. This approach provides a principled mechanism for conservative web agent deployment.
Zitieren
APA
Schmidt, S., Alt, F., & Kinder, J. (2026). Confidence-Aware Agentic RL with Failure Risk Uncertainty Bounds for Web Automation. International Journal of Information and Communication Technology Trends, 6(1), 28-34. https://doi.org/10.66525/ijictt70
Vancouver
Schmidt S, Alt F, Kinder J. Confidence-Aware Agentic RL with Failure Risk Uncertainty Bounds for Web Automation. International Journal of Information and Communication Technology Trends. 2026;6(1):28-34. doi:10.66525/ijictt70
BibTeX
@article{Schmidt2026,
author = {Schmidt, Sebastian and Alt, Florian and Kinder, Johannes},
title = {Confidence-Aware Agentic RL with Failure Risk Uncertainty Bounds for Web Automation},
journal = {International Journal of Information and Communication Technology Trends},
year = {2026},
volume = {6},
number = {1},
pages = {28--34},
doi = {10.66525/ijictt70}
}
EUVIMED