arXiv cs.MA· Amine Andam, Jamal Bentahar, Mustapha Hedabou·· 11 小时前AI 评分32
仅凭观测甚至零访问：针对协作多智能体强化学习的受限黑盒攻击
Constrained Black-Box Attacks Against Cooperative Multi-Agent Reinforcement Learning
AI 导读
研究提出两种受限黑盒攻击设定：攻击者仅能访问智能体观测，或完全无访问权限，通过生成扰动使受害智能体对环境感知错位、无法协作。在22个环境上的实验显示，该类攻击兼具有效性与样本效率。
正文 · 原文
Abstract:Collaborative multi-agent reinforcement learning (MARL) is increasingly deployed in critical domains, yet its vulnerability to adversarial perturbations has not been thoroughly investigated. Most existing adversarial attacks against MARL rely on optimistic assumptions, such as access to policy weights or the ability to train surrogate policies. We investigate whether adversaries with limited capabilities remain effective against MARL agents. In particular, we study two restricted attack settings that are overlooked in the literature: (i) adversaries with access only to agent observations and (ii) adversaries with no access whatsoever. Our approach generates perturbations that intentionally misalign how victim agents perceive their environment, preventing them from collaborating. We conduct extensive experiments across a diverse set of 22 environments. Our results demonstrate both the effectiveness and sample efficiency of our attacks.
| Subjects: | Machine Learning (cs.LG); Multiagent Systems (cs.MA) |
| Cite as: | arXiv:2508.09275 [cs.LG] |
| (or arXiv:2508.09275v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2508.09275 arXiv-issued DOI via DataCite |
Submission history
From: Amine Andam [view email]
[v1]
Tue, 12 Aug 2025 18:31:15 UTC (1,181 KB)
[v2]
Wed, 21 Jan 2026 14:25:32 UTC (1,175 KB)
[v3]
Tue, 3 Nov 2026 21:19:27 UTC (1,235 KB)
来源：arXiv cs.MA · arxiv.org