Robotic Grasp of Language: Unlocking an Open-Ended World for Automation

By blending 2D images with foundation models to build 3D feature fields, a new MIT method helps robots understand and manipulate nearby objects with open-ended…