Event
Flow Control uses general-purpose input to guide VLA actions in real time
1 reports · 1 sources
Jonathan C. Kao and others proposed the flow control method, which writes general-purpose input such as a keyboard into the initial conditions of the flow matching action expert of a vision-language-action (VLA) model to guide robot actions in real time, without retraining or fine-tuning the VLA, and can be used directly. The paper states that because the flow is deterministic, the integration process preserves the user input information while moving toward the trained action distribution, so that the generated actions both reflect user intent and stay close to the VLA's own unguided actions. The paper has been accepted by CoRL 2026.
Report timeline
- Flow Control: Steering VLA Actions in Real Time with Generic Inputs Such as a Keyboard
The paper proposes a flow control method for VLA models that steers robot actions in real time through generic inputs such as a keyboard, without retraining or fine-tuning the VLA. The method writes user input into the initial condition of the VLA's flow matching action expert; because the flow is deterministic, the generated actions both reflect user intent and stay close to the VLA's own unsteered actions.
arXiv cs.RO ·