Premover: Fast Vision-Language-Action Control by Acting Before Instructions Are Complete

Joonha Park, Jiseung Jeong, Taesik Gong

Vision-Language-Action (VLA) policies are typically evaluated as if the user had finished typing or speaking before the robot begins acting. In real deployment, however, users take…