AnyCamVLA: Zero-Shot Camera Adaptation for Viewpoint Robust Vision-Language-Action Models

2026-03-06

AI Summary

This work addresses the sensitivity of vision-language-action (VLA) models to camera viewpoint shifts during deployment, which hinders their adaptability in unstructured environments. The authors propose a zero-shot camera adaptation framework that requires no additional data, fine-tuning, or architectural modifications. By leveraging a feedforward novel-view synthesis model, the method virtually reprojects test-time observations into the training viewpoint in real time, functioning as a plug-and-play module compatible with any RGB-based policy. This approach achieves the first demonstration of fine-tuning-free viewpoint adaptation, significantly outperforming baselines based on data augmentation or 3D features on the LIBERO benchmark. Real-world robotic experiments further validate its robustness to both intrinsic and extrinsic camera parameter variations, as well as to handheld, freely moving cameras.

Technology Category

• Computer Vision: Large Vision Models
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Despite remarkable progress in Vision-Language-Action models (VLAs) for robot manipulation, these large pre-trained models require fine-tuning to be deployed in specific environments. These fine-tuned models are highly sensitive to camera viewpoint changes that frequently occur in unstructured environments. In this paper, we propose a zero-shot camera adaptation framework without additional demonstration data, policy fine-tuning, or architectural modification. Our key idea is to virtually adjust test-time camera observations to match the training camera configuration in real-time. For that, we use a recent feed-forward novel view synthesis model which outputs high-quality target view images, handling both extrinsic and intrinsic parameters. This plug-and-play approach preserves the pre-trained capabilities of VLAs and applies to any RGB-based policy. Through extensive experiments on the LIBERO benchmark, our method consistently outperforms baselines that use data augmentation for policy fine-tuning or additional 3D-aware features for visual input. We further validate that our approach constantly enhances viewpoint robustness in real-world robotic manipulation scenarios, including settings with varying camera extrinsics, intrinsics, and freely moving handheld cameras.

Problem

• Vision-Language-Action models
• camera viewpoint changes
• zero-shot adaptation
• viewpoint robustness
• robot manipulation

Innovation

• zero-shot camera adaptation
• viewpoint robustness
• vision-language-action models
• novel view synthesis
• plug-and-play framework