Lei Han

Existing VLMs misinterpret critical evidence in exploratory manipulation, but a novel distillation approach boosts action recovery accuracy by nearly half.