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

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Wenhao Li, Xueying Jiang, Quanhao Qian, Deli Zhao, Shijian Lu, Gongjie Zhang, Ran Xu
• Venue: arXiv: cs.CV
• Published: 2026-07-06
• Collected: 2026-07-07
• URL: http://arxiv.org/abs/2607.05396v1
• PDF: https://arxiv.org/pdf/2607.05396v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers have developed a new Vision-Language-Action (VLA) model called Camera-Centric VLA (CamVLA). This model aims to improve the performance of existing VLA policies that struggle with camera variations during real-world robot deployment. CamVLA is designed to predict two things: a camera-centric end-effector action and a 6-DoF hand-eye matrix relating cameras to the robot base, allowing the policy to function without explicit knowledge of the camera's position. The model only requires a single monocular RGB image as input and has been shown to improve success rates across diverse viewpoints in both simulation and real-world scenarios.

What the Document Says

• Researchers have developed a new Vision-Language-Action (VLA) model called Camera-Centric VLA (CamVLA).
• CamVLA predicts a camera-centric end-effector action and a 6-DoF hand-eye matrix.
• The model is calibration-free, depth-free, and single-view, requiring only a single monocular RGB image as input.
• Evaluations show that CamVLA improves success rates across diverse unseen viewpoints in both simulation and real-world scenarios.

Key Concepts Mentioned

• Vision-Language-Action (VLA) model
• Camera-Centric VLA (CamVLA)
• camera-centric end-effector action
• 6-DoF hand-eye matrix
• monocular RGB image
• real-world robot deployment

Claims Made by the Document

• CamVLA improves success rates across diverse unseen viewpoints.
• CamVLA is calibration-free, depth-free, and single-view.

Evidence Used

• Evaluations in both simulation and real-world robot data show that CamVLA consistently improves success rates...

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

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/.