JEPA World Models Enhanced with Value-Guided Action Planning
Published:Jan 6, 2026 05:00
•1 min read
•ArXiv ML
Analysis
This paper addresses a critical limitation of JEPA models in action planning by incorporating value functions into the representation space. The proposed method of shaping the representation space with a distance metric approximating the negative goal-conditioned value function is a novel approach. The practical method for enforcing this constraint during training and the demonstrated performance improvements are significant contributions.
Key Takeaways
Reference
“We propose an approach to enhance planning with JEPA world models by shaping their representation space so that the negative goal-conditioned value function for a reaching cost in a given environment is approximated by a distance (or quasi-distance) between state embeddings.”