Learning Temporally and State-Abstracted World Models for Long-Horizon Exploration

William Huang, Benjamin Freed

Keywords: temporal abstraction, state abstraction, hierarchical reinforcement learning, offline reinforcement learning, long horizon behavior

Abstract: Long-horizon robotic exploration requires agents to reason over temporally extended behaviors rather than individual low-level actions. Existing skill-based methods reduce the effective planning horizon but often plan in the original environment state space, while latent world models learn compact representations but typically lack explicit temporal abstraction. We introduce State-Abstracted Latent skills for Temporal planning (SALT), a framework that jointly learns temporal and state abstractions from offline trajectories to support planning in a compact latent space. In preliminary experiments on AntMaze-medium-diverse, an early SALT checkpoint achieves approximately 71.7\% success after only 50 epochs of training, without fine-tuning the world model or planner, yielding performance comparable to established temporally abstract offline planning methods. These results suggest that combining temporal and state abstraction is a promising direction for scalable long-horizon robotic planning.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 20