Paper Overview
Field: Machine Learning Authors: Gaoyue Zhou, Yvonne Wu, Zichen Jeff Cui, Nicolas Ballas, Mido Assran, Lerrel Pinto, Yann LeCun Published: 2026-10-30 arXiv: 2610.42146
Abstract (Full Translation)
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 behavior patterns, ID-WM can synthesize novel long-horizon plans from short-horizon snippets, going beyond the specific behavior patterns seen during training. We validate its effectiveness on three navigation and manipulation suites, showing a 38.5% improvement over state-of-the-art policies and diffusion-based planners. Crucially, by leveraging the hierarchical structure to mitigate compounding errors, our method achieves a 135% improvement over single-level world-model MPC on complex, unseen long-horizon task configurations.
Links
- Paper: https://arxiv.org/abs/2610.42146
- Project page (videos): https://id-wm.github.io