VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Model
This work addresses a key bottleneck in VLA model pretraining for robotics and manipulation tasks, offering a simpler and more effective approach compared to prior methods.
The paper tackled the problem of pretraining Vision-Language-Action policies on internet-scale video, where current methods suffer from appearance bias and information leakage, by introducing VLA-JEPA, a framework that uses leakage-free state prediction in latent space, resulting in consistent gains in generalization and robustness across multiple benchmarks.
Pretraining Vision-Language-Action (VLA) policies on internet-scale video is appealing, yet current latent-action objectives often learn the wrong thing: they remain anchored to pixel variation rather than action-relevant state transitions, making them vulnerable to appearance bias, nuisance motion, and information leakage. We introduce VLA-JEPA, a JEPA-style pretraining framework that sidesteps these pitfalls by design. The key idea is \emph{leakage-free state prediction}: a target encoder produces latent representations from future frames, while the student pathway sees only the current observation -- future information is used solely as supervision targets, never as input. By predicting in latent space rather than pixel space, VLA-JEPA learns dynamics abstractions that are robust to camera motion and irrelevant background changes. This yields a simple two-stage recipe -- JEPA pretraining followed by action-head fine-tuning -- without the multi-stage complexity of prior latent-action pipelines. Experiments on LIBERO, LIBERO-Plus, SimplerEnv and real-world manipulation tasks show that VLA-JEPA achieves consistent gains in generalization and robustness over existing methods.