TempoVLA: Learning Speed-Controllable Vision-Language-Action Policies
2026-06-04 • Robotics
RoboticsArtificial Intelligence
AI summaryⓘ
The authors address how robots need to move quickly during safe parts of a task and slowly during risky, precise moments. They note that past robot models only operate at a single fixed speed, limiting flexibility. The authors created TempoVLA, a system that lets robots change their movement speed on demand by adjusting training data timing and including speed as an input to the model. Their method works well in tests and can automatically speed up or slow down based on task difficulty when combined with a large multimodal model.
robot manipulationvision-language-action modelsvariable-speed controltrajectory augmentationspeed conditioningpolicy modelrobot motion planningmultimodal modelsdata augmentation
Authors
Dong Jing, Jingchen Nie, Tianqi Zhang, Jiaqi Liu, Huaxiu Yao, Zhiwu Lu, Mingyu Ding
Abstract
Robot manipulation alternates between low-risk transit phases that call for fast execution and high-risk contact stages that demand slow, precise motion. Yet existing Vision-Language-Action models (VLAs) only inherit a single fixed speed from training demonstrations. Prior efforts to accelerate VLAs through model compression, KV-cache reuse, or reinforcement learning only shift the policy from one fixed speed to another, and leave deceleration almost unexplored. We observe that the magnitude of each predicted action already governs how fast the robot moves, opening a direct route to controllable execution speed. We turn this observation into TempoVLA, a single VLA whose execution speed is controlled by an explicit condition. TempoVLA combines two coupled components. (1) A data-side Variable-Speed Trajectory Augmentation (VSTA) that re-times demonstration to any target speed by merging or splitting actions while preserving its motion semantics. (2) A model-side conditioning mechanism that feeds the speed to the policy. Statistics show that VSTA reaches the requested speed with negligible motion error. Experiments in simulation and on real-world tasks demonstrate that TempoVLA achieves flexible speed control in both directions, while VSTA additionally boosts the default $1\times$ performance via better data utilization. Furthermore, by cooperating with a large multimodal model, TempoVLA realizes dynamic speed control, accelerating through low-risk phases and decelerating for high-risk ones.