Completion at the Boundary (CaB): Deployable Switching with Completion-Aware Control under Limited Calibration
Researchers have introduced Completion at the Boundary (CaB), a novel method for vision-language-action (VLA) agents to determine when a task is complete. This is crucial for composite instructions like "do A, then B," where mistimed handoffs can lead to failures. CaB predicts completion objects using Boundary-Phase Tokens and converts these into auditable switching decisions, improving execution quality on a Minecraft VLA benchmark. AI
IMPACT Improves AI agent reliability in executing sequential instructions, crucial for complex task automation.