Using language to give robots a better grasp of an open-ended world
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 language prompts.