arXiv：cs.AI· Brandon Gary Kaplowitz, Osaze James Obahor, Christian Schroeder de Witt·· 3 小时前
MA-JEPA：面向多智能体强化学习的联合嵌入世界模型
MA-JEPA: Joint-Embedding World Models for Multi-Agent Reinforcement Learning
AI 导读
MA-JEPA 是一种随机世界模型，用目标表征预测替代观测重建，实现集中训练、分散执行的多智能体强化学习。它通过类别潜状态与因果 Transformer，结合后验和动作条件动力学预测目标进行训练，再用潜在想象做 actor-critic 学习。在 SMAC 的 8 张评测地图中，有 4 张的平均胜率达到或超过已报告的最强对比方法。
正文
Abstract:World models improve sample efficiency by training policies on imagined trajectories, but their usefulness depends on learning representations that capture the information needed for future control. We study whether self-supervised joint-embedding prediction (JEPA) can provide this learning signal for multi-agent reinforcement learning. We introduce MA-JEPA, a stochastic world model that replaces observation reconstruction with prediction of target representations, enabling model-based multi-agent reinforcement learning with centralized training and decentralized execution. A categorical latent state and a causal Transformer are trained with posterior and action-conditioned dynamics prediction objectives and are then used for actor-critic learning from latent imagination. A training-only joint predictor conditions on all agents' local states and actions to predict each agent's next local observation embedding. These predictions are passed through the same local posterior used during real interaction with a centralized critic that is used only for value learning, with execution remaining decentralized. Our experiments show that this architecture performs strongly on SMAC, matching or exceeding the strongest reported comparator mean win rate on four of eight evaluated maps.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA) |
| Cite as: | arXiv:2609.33563 [cs.LG] |
| (or arXiv:2609.33563v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2609.33563 arXiv-issued DOI via DataCite |
Submission history
From: Brandon Kaplowitz [view email]
[v1]
Sun, 27 Sep 2026 13:37:30 UTC (2,021 KB)
[v2]
Thu, 8 Oct 2026 17:57:30 UTC (2,022 KB)
来源：arXiv：cs.AI · arxiv.org