Premover: Fast Vision-Language-Action Control by Acting Before Instructions Are Complete
arXiv:2605.12160v1 Announce Type: new
Abstract: 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 several seconds to ent…