World Model-Based Planning
Overview
World model-based planning studies how an agent can predict the consequences of actions and use those predictions to reach goals, avoid risks, or coordinate with others. Models may predict images, compact latent states, object relations, or physical dynamics, and planners use methods such as model-predictive control, CEM, tree search, and learned action policies. Work spans robot manipulation and navigation, autonomous driving, games, and language-guided agents, with approaches including hierarchical subgoals, uncertainty estimates, and symbolic or physics-based structure. A central challenge is that accurate-looking or low-error predictions may still rank actions poorly, especially over long horizons or under distribution shift, so researchers are developing planning-focused representations, diagnostics, and more efficient closed-loop adaptation.
Latest papers
308 papers44 papers in the last four weeks, down 45% on the four weeks before. 0.3% of all new papers.
Do Enterprise Systems Need Learned World Models? The Importance of Context to Infer Dynamics
The DAWN of World-Action Interactive Models
Latent Geometry Beyond Search: Amortizing Planning in World Models
Is the Future Compatible? Diagnosing Dynamic Consistency in World Action Models
Predictive but Not Plannable: RC-aux for Latent World Models
See Tomorrow, Act Today: Foresight-Driven Autonomous Driving
what will happen next" before acting. We introduce ForeSight, a foundation world model centric planning framework that reframes autonomous driving as anticipatory decision-making. Rather than treating world models as auxiliary components, ForeSight makes future scene imagination the primary driver of action prediction. Our approach operates in two stages: (1) generating plausible future visual worlds via a pretrained world model, and (2) planning actions conditioned on these imagined futures. This paradigm shift from what should I do now?" to ``what will happen, and how should I respond?" enables genuinely anticipatory rather than reactive planning. By grounding decisions in anticipated contexts rather than present observations alone, ForeSight navigates dynamic, interactive scenarios more effectively. Extensive experiments on NAVSIM and nuScenes demonstrate that explicit future imagination significantly outperforms previous state-of-the-art alternatives, validating our foresight-driven approach.