arXiv Robotics — research abstracts· Boyao Han, Chen Shi, Jingjing Qian, ZhuoTan Tian, Li Jiang·· 10 days agoEditorial score63
VersaCamVLA: Camera-Configurable VLA Policies for Robotic Manipulation
VersaCamVLA: Camera-Configurable VLA Policies for Robotic Manipulation
Summary
The paper presents VersaCamVLA, a framework that decouples camera-set representation from action learning, allowing for flexible and robust robotic manipulation across varying camera configurations. It uses multi-signal target-view prediction and Wrist-Augmented Pose Sampling to generate diverse poses, and injects compact scene tokens into a pretrained VLA model for deployment.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
Source：arXiv Robotics — research abstracts · arxiv.org