Uncertainty-Aware Policy Optimization for Robust Multi-Agent Decision Making under Partial Observability
Fundamental Scientific Reports in Multidisciplinary Areas · 01.04.2026 · 2 (1) · 66-73
DOI: 10.66238/fsrma56 ↗Kurz gefasst
Fachartikel in der Zeitschrift Fundamental Scientific Reports in Multidisciplinary Areas, online erschienen am 01.04.2026. Verfasst von Erik Andersson und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Decision-making in partially observable environments introduces uncertainty that can degrade coordination performance. This study develops an uncertainty-aware reinforcement learning framework by integrating soft actor-critic (SAC) with entropy regularization and Bayesian value estimation. The approach explicitly models uncertainty in both state representation and reward estimation, enabling more stable policy updates. Evaluation is performed on 10,000 tasks with varying levels of observation noise and incomplete information, including negotiation and multi-stage planning scenarios. The proposed method improves success rate from 69.2% to 81.5% and reduces variance in cumulative reward by 27.4% compared to standard SAC. Additionally, calibration metrics indicate better alignment between predicted and actual outcomes. The results demonstrate that uncertainty-aware optimization enhances robustness in collaborative decision systems.
Zitieren
APA
Andersson, E., Svensson, J., & Johansson, L. (2026). Uncertainty-Aware Policy Optimization for Robust Multi-Agent Decision Making under Partial Observability. Fundamental Scientific Reports in Multidisciplinary Areas, 2(1), 66-73. https://doi.org/10.66238/fsrma56
Vancouver
Andersson E, Svensson J, Johansson L. Uncertainty-Aware Policy Optimization for Robust Multi-Agent Decision Making under Partial Observability. Fundamental Scientific Reports in Multidisciplinary Areas. 2026;2(1):66-73. doi:10.66238/fsrma56
BibTeX
@article{Andersson2026,
author = {Andersson, Erik and Svensson, Johan and Johansson, Lars},
title = {Uncertainty-Aware Policy Optimization for Robust Multi-Agent Decision Making under Partial Observability},
journal = {Fundamental Scientific Reports in Multidisciplinary Areas},
year = {2026},
volume = {2},
number = {1},
pages = {66--73},
doi = {10.66238/fsrma56}
}
EUVIMED