arXiv cs.AI· Raz Shmueli, Shahaf S. Shperberg·· 2 天前AI 评分22
CA-MAMBA：在想象中进行受约束多智能体强化学习
Constrained Multi-Agent Reinforcement Learning in Imagination
AI 导读
CA-MAMBA 将多智能体世界模型中的成本预测移入想象空间，在 SMAC 中位配置上所需交互次数仅为无模型基线的 1/10 至 1/20，在同等预算下 MAMuJoCo 奖励为基线的 2.8 至 12 倍。在每基准单任务的受控测试中，将模型成本优势按平价定价可使成本盲学习者的成本降低一半以上，并满足学习型乘数未能达到的严格连续控制限制，表明模型成本估计并非瓶颈。
正文 · 原文
Abstract:Predicting cost in a multi-agent world model moves most constrained policy updates into imagination: the stronger model-free baseline needs $10$ to $20$ times CA-MAMBA's interactions on the median SMAC configuration, and CA-MAMBA earns $2.8$ to $12$ times the baselines' MAMuJoCo reward at equal budget. In controlled tests on one task per benchmark, pricing the model's cost advantages at parity more than halves a cost-blind learner's cost and meets a tight continuous-control limit the learned multiplier misses, suggesting the model's cost estimates are not the bottleneck.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.21319 [cs.AI] |
| (or arXiv:2610.21319v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.21319 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shahaf S. Shperberg [view email]
[v1]
Thu, 15 Oct 2026 06:09:04 UTC (1,307 KB)
来源：arXiv cs.AI · arxiv.org