ThinkJEPA: Empowering Latent World Models with Large Vision-Language Reasoning Model

TL;DR

This paper introduces a VLM-guided latent world modeling framework that combines dense motion prediction with semantic reasoning, improving long-horizon forecasting in video-based tasks.

Contribution

It proposes a dual-temporal pathway model integrating dense JEPA dynamics with a large-scale vision-language model for semantic guidance, enhancing long-term prediction accuracy.

Findings

Outperforms baseline models in hand-manipulation trajectory prediction

Achieves more robust long-horizon rollout behavior

Effectively integrates semantic guidance into dense motion modeling

Abstract

Recent progress in latent world models (e.g., V-JEPA2) has shown promising capability in forecasting future world states from video observations. Nevertheless, dense prediction from a short observation window limits temporal context and can bias predictors toward local, low-level extrapolation, making it difficult to capture long-horizon semantics and reducing downstream utility. Vision--language models (VLMs), in contrast, provide strong semantic grounding and general knowledge by reasoning over uniformly sampled frames, but they are not ideal as standalone dense predictors due to compute-driven sparse sampling, a language-output bottleneck that compresses fine-grained interaction states into text-oriented representations, and a data-regime mismatch when adapting to small action-conditioned datasets. We propose a VLM-guided JEPA-style latent world modeling framework that combines dense-frame dynamics modeling with long-horizon semantic guidance via a dual-temporal pathway: a dense JEPA branch for fine-grained motion and interaction cues, and a uniformly sampled VLM \emph{thinker} branch with a larger temporal stride for knowledge-rich guidance. To transfer the VLM's progressive reasoning signals effectively, we introduce a hierarchical pyramid representation extraction module that aggregates multi-layer VLM representations into guidance features compatible with latent prediction. Experiments on hand-manipulation trajectory prediction show that our method outperforms both a strong VLM-only baseline and a JEPA-predictor baseline, and yields more robust long-horizon rollout behavior.