Robots halt start-stop syndrome: VLASH enables continuous motion

Researchers have developed a system called VLASH to eliminate the "start-stop syndrome" in robots powered by vision-language-action (VLA) models. While traditional VLA systems move intermittently—pausing to process each command before executing the next movement—VLASH enables continuous motion by implementing parallel planning. The system allows a robot to calculate the trajectory of its next action while the motors are still executing the current command. To reduce the computational burden, the model does not predict the entire complex dynamics of the surrounding environment; instead, it focuses solely on predicting the future state of the robot itself. This optimization enables the system to operate using standard hardware, such as a gaming laptop equipped with an RTX 5090 graphics card. Parameter VLASH Result Latency reduction (realistic tests) Up to 11.8x Acceleration on slow hardware Over 30x Model fine-tuning speed 3.26x faster Testing and Performance The system was tested by specialists from MIT, Nvidia, Caltech, and other leading universities. The robots performed tasks involving sorting, stacking, and playing table tennis. Using various hardware accelerators to measure response times, the team found that latency was reduced by more than 11 times compared to traditional sequential logic.