English

cVLA: Towards Efficient Camera-Space VLAs

Robotics 2025-12-23 v2 Machine Learning

Abstract

Vision-Language-Action (VLA) models offer a compelling framework for tackling complex robotic manipulation tasks, but they are often expensive to train. In this paper, we propose a novel VLA approach that leverages the competitive performance of Vision Language Models (VLMs) on 2D images to directly infer robot end-effector poses in image frame coordinates. Unlike prior VLA models that output low-level controls, our model predicts trajectory waypoints, making it both more efficient to train and robot embodiment agnostic. Despite its lightweight design, our next-token prediction architecture effectively learns meaningful and executable robot trajectories. We further explore the underutilized potential of incorporating depth images, inference-time techniques such as decoding strategies, and demonstration-conditioned action generation. Our model is trained on a simulated dataset and exhibits strong sim-to-real transfer capabilities. We evaluate our approach using a combination of simulated and real data, demonstrating its effectiveness on a real robotic system.

Keywords

Cite

@article{arxiv.2507.02190,
  title  = {cVLA: Towards Efficient Camera-Space VLAs},
  author = {Max Argus and Jelena Bratulic and Houman Masnavi and Maxim Velikanov and Nick Heppert and Abhinav Valada and Thomas Brox},
  journal= {arXiv preprint arXiv:2507.02190},
  year   = {2025}
}

Comments

20 pages, 10 figures; CoRL 2025 Workshop on Generalizable Priors for Robot Manipulation

R2 v1 2026-07-01T03:44:06.095Z