AgentOWL：联合学习分层神经 options 与抽象世界模型
Joint Learning of Hierarchical Neural Options and Abstract World Model
AI 导读
研究 team 提出 AgentOWL 方法，以样本高效的方式联合学习抽象世界模型（同时在状态和时间上做抽象）与一组分层神经 options。在部分 Object-Centric Atari 游戏上，该方法比基线用更少数据学到更多技能，并具备基线所没有的学习与泛化能力。论文已被 NeurIPS 2026 接收。
正文 · 原文
Abstract:Building agents that can perform new skills by composing existing skills is a long-standing goal of AI agent research. Towards this end, we investigate how to efficiently acquire a sequence of skills, formalized as hierarchical neural options. However, existing model-free hierarchical reinforcement algorithms need a lot of data. We propose a novel method, which we call AgentOWL (Option and World model Learning Agent), that jointly learns -- in a sample efficient way -- an abstract world model (abstracting across both states and time) and a set of hierarchical neural options. We show, on a subset of Object-Centric Atari games, that our method can learn more skills using less data than baseline methods and possesses learning and generalization capabilities that the baselines do not have.
| Comments: | Accepted to NeurIPS 2026 |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2602.02799 [cs.LG] |
| (or arXiv:2602.02799v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2602.02799 arXiv-issued DOI via DataCite |
Submission history
From: Wasu Top Piriyakulkij [view email]
[v1]
Mon, 2 Feb 2026 20:58:11 UTC (3,094 KB)
[v2]
Mon, 11 May 2026 21:28:20 UTC (1,860 KB)
[v3]
Tue, 10 Nov 2026 21:39:16 UTC (1,974 KB)
来源：arXiv cs.AI · arxiv.org