arXiv cs.AI· Gaoyue Zhou, Yvonne Wu, Zichen Jeff Cui, Nicolas Ballas, Mido Assran, Lerrel Pinto, Yann LeCun·· 4 小时前AI 评分26
ID-WM：基于隐式动力学的分层世界模型
Hierarchical World Models with Implicit Dynamics
AI 导读
研究者提出 ID-WM（Implicit Dynamics World Models），一种将状态空间可达性与局部控制解耦的分层世界模型：高层模型在状态流形上采样可行转移，低层世界模型作为逆动力学求解器将其落地为精确动作。该方法在三个导航与操作测试套件中比 SOTA 策略和扩散规划器提升 38.5%，相比单层世界模型 MPC 提升 135%，尤其在长时程任务的未见复杂配置中缓解了累积误差问题。
正文 · 原文
Abstract:Planning with world models enables test-time adaptation for embodied agents, offering superior generalization over direct policy learning. However, long-horizon tasks remain hindered by cumulative errors and search complexity. Existing hierarchical methods often use rigid macro-actions or latent skills, leading to brittle representations and limited fine-grained control. We introduce ID-WM (Implicit Dynamics World Models), a hierarchical world model that decouples state-space reachability from local control. Our framework utilizes a high-level model to sample feasible transitions across the state manifold and a low-level world model as an inverse dynamics solver to ground these transitions into precise actions. This allows the planner to identify reachable subgoals without being constrained by low-level execution details. By prioritizing local physical feasibility over global behavioral patterns, ID-WM synthesizes novel long-horizon plans from short segments, transcending the specific behavioral modes seen during training. We demonstrate its effectiveness across three navigation and manipulation suites, achieving a 38.5% improvement over state-of-the-art policy and diffusion-based planners. Crucially, by leveraging this hierarchical structure to mitigate compounding errors, our approach yields a 135% performance gain over single-level world model MPC, particularly in complex, unseen configurations for long-horizon tasks. Videos can be viewed at this https URL
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.42146 [cs.RO] |
| (or arXiv:2610.42146v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.42146 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Gaoyue Zhou [view email]
[v1]
Fri, 30 Oct 2026 17:59:25 UTC (8,888 KB)
来源：arXiv cs.AI · arxiv.org