“The Gemini Robotics-ER model's capabilities include object detection, pointing, trajectory and grasp prediction, multi-view correspondence, and 3D bounding box predictions.”