Problem
Research questions and friction points this paper is trying to address.
VLA deployment
real-time execution
robotic dexterity
execution speed
real-world robotics
Realtime-VLA V2: Learning to Run VLAs Fast, Smooth, and Accurate
🤖 AI Summary
Deploying vision-language-action (VLA) models on real robots often entails a trade-off among execution speed, accuracy, and dexterity. This work proposes a synergistic optimization framework that integrates system calibration, motion planning, and learning-based control, enabling the first real-time, end-to-end VLA execution on a lightweight robotic arm at speeds approaching both human performance and hardware limits. By automatically identifying the optimal execution velocity through learning and combining it with high-precision trajectory planning and comprehensive system calibration, the deployed system achieves human-level fluency and accuracy in real-world tasks. Experimental validation using unaccelerated video demonstrations and inferred trajectories confirms the system’s performance, substantially advancing the practical deployment boundary of VLA models on physical robots.
📝 Abstract
In deployment of the VLA models to real-world robotic tasks, execution speed matters. In previous work arXiv:2510.26742 we analyze how to make neural computation of VLAs on GPU fast. However, we leave the question of how to actually deploy the VLA system on the real robots open. In this report we describe a set of practical techniques to achieve the end-to-end result of running a VLA-driven robot at an impressive speed in real world tasks that require both accuracy and dexterity. The stack of technology ranges across calibration, planning & control, and learning based method to identify optimal execution speed. In the tasks we show, the robot even executes in a speed on par with casual human operation and approaching the hardware limit of our lightweight arm. The unaccelerated videos and inference traces are provided in https://dexmal.github.io/realtime-vla-v2/.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Realtime-VLA
robotic deployment
end-to-end execution
optimal execution speed
vision-language-action
🔎 Similar Papers
No similar papers found.
