Vision-Language-Action (VLA) models have made significant strides in visuomotor policy learning but fundamentally remain reactive, mapping current observations and language to actions without explicit forward prediction of world dynamics. Existing visual foresight methods can predict future visual states but lack explicit motion guidance: they show where to go but not how to get there.
We argue that future feature prediction and sparse point tracking are naturally complementary: the former provides the goal state while the latter captures the continuous motion path toward it. To address this, we propose the FoMoVLA framework, which augments VLA representations with explicit spatio-temporal supervision by jointly learning future feature foresight and sparse 2D point tracking, thereby enhancing the continuous action policy.
FoMoVLA introduces compact foresight tokens to decode future feature states and decodes sparse temporal 2D point trajectories to model compact geometric motion. Both are coupled through a lightweight future-conditioned cross-attention module that enables consistent reasoning between anticipated states and point dynamics. Extensive experiments on LIBERO, RoboCasa GR-1 Tabletop, and LIBERO-Plus demonstrate state-of-the-art performance and strong zero-shot generalization. The project page is available at FoMoVLA Project.
Blogger's Review: FoMoVLA significantly enhances the decision-making capability of VLA models in complex environments by integrating visual foresight with motion guidance. This approach not only improves the model's reasoning ability but also broadens the scope for future research, especially in dynamic scene applications. Its outstanding performance across multiple benchmark datasets validates the framework's effectiveness and practicality.