From Spatial to Actions: Grounding Vision-Language-Action Model in Spatial Foundation Priors
This addresses the problem of limited generalization and adaptability in 3D real-world actions for robotics and AI systems, representing a novel paradigm rather than an incremental improvement.
The paper tackles the spatial reasoning gap in vision-language-action models by introducing FALCON, which injects 3D spatial tokens into the action head using spatial foundation models from RGB alone, achieving state-of-the-art performance across simulation benchmarks and real-world tasks.
Existing vision-language-action (VLA) models act in 3D real-world but are typically built on 2D encoders, leaving a spatial reasoning gap that limits generalization and adaptability. Recent 3D integration techniques for VLAs either require specialized sensors and transfer poorly across modalities, or inject weak cues that lack geometry and degrade vision-language alignment. In this work, we introduce FALCON (From Spatial to Action), a novel paradigm that injects rich 3D spatial tokens into the action head. FALCON leverages spatial foundation models to deliver strong geometric priors from RGB alone, and includes an Embodied Spatial Model that can optionally fuse depth, or pose for higher fidelity when available, without retraining or architectural changes. To preserve language reasoning, spatial tokens are consumed by a Spatial-Enhanced Action Head rather than being concatenated into the vision-language backbone. These designs enable FALCON to address limitations in spatial representation, modality transferability, and alignment. In comprehensive evaluations across three simulation benchmarks and eleven real-world tasks, our proposed FALCON achieves state-of-the-art performance, consistently surpasses competitive baselines, and remains robust under clutter, spatial-prompt conditioning, and variations in object scale and height.