PACT-VLA: Speculative Action-Chunk Verification with Pre-Commit Observations

Vision-language-action (VLA) policies are commonly deployed with asynchronous inference in real-world settings, where the next action chunk is generated while the robot is still executing the previous one. By the time the new chunk is committed, the robot and scene have changed, so the chunk is conditioned on a stale observation and task success drops sharply as this inference delay grows.

From the abstract. Our summary is in progress.