Back to AI News
arxiv

Fast-JEPA speeds up latent world model planning with direct candidate outcome prediction

Fast-JEPA is a latent world model planning framework that reorganizes predictive computation to avoid recursive state rollout when comparing candidate action sequences. It builds a goal-directed plan to guide search and directly predicts the outcomes of complete candidates, targeting a better success-latency tradeoff. Across PushT, Reacher, Cube and TwoRoom the authors report 96.50% average success with planning 2.41x faster than INTACT, with gains holding across search-budget sweeps. The paper also introduces DynaGoal, a benchmark for world-model planning under dynamic out-of-distribution conditions, execution deadlines and controlled response delays, where Fast-JEPA reports 73.83% and 87.67% average success on the Shift and Deadline variants.

AI Brief

It offers a concrete planning-cost reduction with reported success rates and a new benchmark for latency-constrained world-model planning, useful to researchers working on model-based control.

arXiv:2611.25927v1 Announce Type: cross Abstract: Latent world models predict the outcomes of action sequences, but comparing candidate plans through recursive rollout remains expensive at test time. We introduce Fast-JEPA, a latent world model planning framework that advances the success-latency Pareto frontier of predictive search. Fast-JEPA reorganizes predictive computation to support effective action-sequence search at low planning cost. It constructs a goal-directed plan to guide search and directly predicts the outcomes of complete candidates for efficient comparison, avoiding recursive state rollout during refinement. Across PushT, Reacher, Cube, and TwoRoom, Fast-JEPA achieves 96.50% average success rate with 2.41x faster planning than INTACT. Search-budget sweeps show that its competitive success and low latency extend across candidate counts and refinement depths. We further introduce DynaGoal, a benchmark for evaluating world-model planning under dynamic out-of-distribution conditions, execution deadlines, and controlled response delays. Fast-JEPA achieves average success rates of 73.83% on DynaGoal-Shift and 87.67% on DynaGoal-Deadline, demonstrating the practical value of low-latency predictive planning under changing control conditions and execution-time constraints. Our code is available at https://github.com/Sofilyzia/Fast-JEPA.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!