Paper 2510.12276
Spatial Forcing: Implicit Spatial Representation Alignment for Vision-language-action Model
- Published
- Oct 2025
- Research lab
- Independent
- Citations
- 93
- GitHub
- 277 stars
01 In brief
Summary
The paper introduces Spatial Forcing (SF), a method to enhance the spatial awareness of Vision-Language-Action (VLA) models without explicit 3D inputs.
VLA models, built on 2D-pretrained VLMs, lack 3D understanding, limiting their robotic manipulation performance.
Existing solutions using depth sensors or point clouds face issues like sensor noise and data heterogeneity, while depth estimators are suboptimal.
SF aligns intermediate visual embeddings of VLAs with spatial representations from a pretrained 3D foundation model (VGGT) using cosine similarity, implicitly forcing the model to learn 3D comprehension.
Experiments on LIBERO and RoboTwin benchmarks show SF achieves state-of-the-art success rates (98.5% average on LIBERO), surpassing both 2D and explicit 3D VLAs.
SF also accelerates training by up to 3.8x and improves data efficiency, achieving 75.8% success with only 5% of training data.
Real-world experiments on a bimanual platform demonstrate higher success rates across tasks with limited demonstrations (40 for single-arm, 20 for bimanual), confirming its practical applicability and data efficiency.
The method adds no inference overhead and is compatible with existing VLA architectures.
02 From the paper
Abstract
Vision-language-action (VLA) models have recently shown strong potential in enabling robots to follow language instructions and execute precise actions. However, most VLAs are built upon vision-language models pretrained solely on 2D data, which lack accurate spatial awareness and hinder their ability to operate in the 3D physical world. Existing solutions attempt to incorporate explicit 3D sensor inputs such as depth maps or point clouds, but these approaches face challenges due to sensor noise, hardware heterogeneity, and incomplete depth coverage in existing datasets. Alternative methods that estimate 3D cues from 2D images also suffer from the limited performance of depth estimators. We propose Spatial Forcing (SF), a simple yet effective alignment strategy that implicitly forces VLA models to develop spatial comprehension capabilities without relying on explicit 3D inputs or depth estimators. SF aligns intermediate visual embeddings of VLAs with geometric representations produced by pretrained 3D foundation models. By enforcing alignment at intermediate layers, SF guides VLAs to encode richer spatial representations that enhance action precision. Extensive experiments in simulation and real-world environments demonstrate that SF achieves state-of-the-art results, surpassing both 2D- and 3D-based VLAs. SF further accelerates training by up to 3.8x and improves data efficiency across diverse robotic tasks. Project page is at https://spatial-forcing.github.io/