embodied-r1-robotic-reasoning

Bridge vision-to-action gap using pointing as unified intermediate representation, enabling 56.2% success on manipulation tasks without task-specific fine-tuning.

Independent GitHub snapshot · Automated classification · Original documentation governs