arXiv cs.MA· Chengxi Zhong, Yongzhe Chang·· 11 小时前AI 评分27
Anchor and Perturb（AnP）：用探索注入修复多智能体协调中的懒惰智能体
Anchor and Perturb: Lazy Agent Remediation by Exploration Injection
AI 导读
Anchor and Perturb（AnP）是一个轻量框架，通过将探索性方差注入与循环流形稳定性解耦来修复多智能体协调失败。它隔离表现不佳的懒惰智能体并向目标坐标注入非对称探索脉冲，同时将已收敛队友锚定在贪心利用上。基准测试显示，AnP 可将崩溃的联合策略从 5% 评估胜率低谷恢复至 85%，并维持 90% 的峰值胜率，且无需修改网络结构。
正文 · 原文
Abstract:Anchor and Perturb (AnP) is a lightweight framework that resolves multi-agent coordination failures by decoupling exploratory variance injection from recurrent manifold stability. Existing remediation strategies predominantly alter mixing network architectures or enforce simultaneous exploration across the collective, which inevitably precipitates severe temporal-difference penalties in non-monotonic reward spaces. Specifically, AnP isolates underperforming lazy agents and injects an asymmetric exploratory pulse into targeted coordinates whilst anchoring converged teammates to nominal greedy exploitation. Empirical telemetry benchmarks demonstrate that AnP successfully rescues collapsed joint policies (recovering from a 5% evaluation win rate nadir back to 85%) and facilitates escape from suboptimal coordination plateaus, sustaining peak win rates of 90% without requiring structural network modifications.
| Comments: | 12 pages, 4 figures; v2 includes multi-cycle automated benchmarks and Section 5.3 additions |
| Subjects: | Multiagent Systems (cs.MA) |
| Cite as: | arXiv:2609.27365 [cs.MA] |
| (or arXiv:2609.27365v2 [cs.MA] for this version) | |
| https://doi.org/10.48550/arXiv.2609.27365 arXiv-issued DOI via DataCite |
Submission history
From: Chengxi Zhong [view email]
[v1]
Wed, 23 Sep 2026 05:08:37 UTC (37 KB)
[v2]
Fri, 30 Oct 2026 07:54:43 UTC (428 KB)
来源：arXiv cs.MA · arxiv.org