Model-Based Planning
Overview
Model-based planning uses a model of how actions change the world to look ahead before acting, supporting decisions in robotics, navigation, games, tool use, and scheduling. Common approaches include model predictive control, Monte Carlo tree search and dynamic programming, symbolic action models, and learned latent or generative world models; language-model agents increasingly combine these with search, constraints, and execution feedback. Benchmarks range from simulated continuous control and manipulation to partially observable tasks, dynamic environments, and real-robot trials. A central challenge is that good predictions do not necessarily yield good plans: models must rank actions reliably over useful horizons while controlling computation, adapting to uncertainty and distribution shifts, and avoiding plans that exploit model errors.
Latest papers
222 papers19 papers in the last four weeks, down 55% on the four weeks before. 0.1% of all new papers.
Reactive Motion Generation With Particle-Based Perception in Dynamic Environments
When Remembering and Planning are Worth it: Navigating under Change
NavDreamer: Video Models as Zero-Shot 3D Navigators
Textual Planning with Explicit Latent Transitions
Imagine-then-Plan: Agent Learning from Adaptive Lookahead with World Models
ITP), a unified framework for agent learning via lookahead imagination, where an agent's policy model interacts with the learned world model, yielding multi-step ``imagined'' trajectories. Since the imagination horizon may vary by tasks and stages, we introduce a novel adaptive lookahead mechanism by trading off the ultimate goal and task progress. The resulting imagined trajectories provide rich signals about future consequences, such as achieved progress and potential conflicts, which are fused with current observations, formulating a partially observable and imaginable Markov decision process to guide policy learning. We instantiate ITP with both training-free and reinforcement-trained variants. Extensive experiments across representative agent benchmarks demonstrate that ITP significantly outperforms competitive baselines. Further analyses validate that our adaptive lookahead largely enhances agents' reasoning capability, providing valuable insights into addressing broader, complex tasks. Our code and data will be publicly available at https://github.com/loyiv/ITP.