From Fixed to Free Cameras: Calibration-Free View-Robust Vision-Language-Action Model

2026-07-06

AI Summary

This work addresses the challenge that existing vision-language-action (VLA) models rely on explicit extrinsic camera calibration and struggle to generalize under varying camera viewpoints in real-world settings. The authors propose CamVLA, the first end-to-end viewpoint-robust VLA model that operates without camera calibration or depth information. CamVLA decouples camera geometry from manipulation control by predicting end-effector actions in the camera-centered coordinate frame directly from monocular RGB images, then leverages an estimated 6-DoF hand-eye transformation matrix to deterministically map these actions into the robot base frame. Experiments demonstrate that CamVLA significantly improves task success rates across both simulated and real robotic environments under previously unseen viewpoints, exhibiting strong generalization and practical deployment capabilities.

Abstract

Real-world robot deployment rarely maintains the training-stage camera setup, where cameras often experience repositioning or remounting depending on actual scenarios. Existing view-robust Vision-Language-Action (VLA) policies tolerate such camera variations only when the camera extrinsics are explicitly provided, making them fragile and hard to use especially when view robustness is critical. We argue that the policy should not be told where the camera is, but rather figure it out by itself. To this end, we introduce Camera-Centric VLA (CamVLA), a new VLA model that decouples manipulation controls from camera geometry by predicting (i) a camera-centric end-effector action expressed in the local camera frame, and (ii) a 6-DoF hand-eye matrix relating cameras to the robot base. A deterministic geometric transformation composes the two predictions into a robot base-frame action. This disentangles how I should move in pose-independent camera-centric action generation from where I am looking from in camera-perspective geometric grounding. The resulting policy is calibration-free, depth-free, and single-view, requiring only a single monocular RGB image as the visual observation and task instruction at deployment. Evaluations in both simulation and real-world robot data show that CamVLA consistently improves success rates across diverse unseen viewpoints. Project page: https://alibaba-damo-academy.github.io/CamVLA/.

Problem

• view robustness
• camera extrinsics
• Vision-Language-Action
• robot deployment
• calibration-free

Innovation

• calibration-free
• view-robust
• vision-language-action
• camera-centric action
• hand-eye transformation