●builderYou can use the open-source Refiner implementation to automate subtask annotation for robot training at a significantly lower cost than human labeling.
●researcherThis benchmark provides a standardized metric for evaluating how well vision-language models decompose high-level instructions into actionable robotic segments.