ID-WM decouples reachability from control in hierarchical world models
Original title: Hierarchical World Models with Implicit Dynamics
ID-WM uses high-level sampling of feasible transitions and low-level inverse dynamics solving to enable scalable long-horizon planning with 38.5% better performance.
Planning with world models allows test-time adaptation for embodied agents but suffers from cumulative errors in long-horizon tasks. ID-WM decouples state-space reachability from local control using a high-level model and low-level inverse dynamics solver. It synthesizes novel plans from short segments, achieving gains over state-of-the-art policy and diffusion-based planners.
- High-level model samples feasible transitions across state manifold
- Low-level model acts as inverse dynamics solver
- 38.5% improvement over SOTA policy and diffusion planners
- 135% gain over single-level world model MPC in complex configurations
From the original article
1 min read at arxiv.orgPlanning 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.
By Gaoyue Zhou · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.