Grounding Actions in Camera Space: Observation-Centric Vision-Language-Action Policy

TL;DR

This paper introduces OC-VLA, a framework that grounds vision-language-action models directly in camera space, improving generalization and robustness across different viewpoints in robotic manipulation tasks.

Contribution

The paper proposes a novel observation-centric approach that transforms action predictions into camera space using extrinsic calibration, enhancing model generalization without major architecture changes.

Findings

Accelerates convergence in robotic tasks

Improves success rates across viewpoints

Enhances cross-view generalization

Abstract

Vision-Language-Action (VLA) models frequently encounter challenges in generalizing to real-world environments due to inherent discrepancies between observation and action spaces. Although training data are collected from diverse camera perspectives, the models typically predict end-effector poses within the robot base coordinate frame, resulting in spatial inconsistencies. To mitigate this limitation, we introduce the Observation-Centric VLA (OC-VLA) framework, which grounds action predictions directly in the camera observation space. Leveraging the camera's extrinsic calibration matrix, OC-VLA transforms end-effector poses from the robot base coordinate system into the camera coordinate system, thereby unifying prediction targets across heterogeneous viewpoints. This lightweight, plug-and-play strategy ensures robust alignment between perception and action, substantially improving model resilience to camera viewpoint variations. The proposed approach is readily compatible with existing VLA architectures, requiring no substantial modifications. Comprehensive evaluations on both simulated and real-world robotic manipulation tasks demonstrate that OC-VLA accelerates convergence, enhances task success rates, and improves cross-view generalization. The code will be publicly available.