Carnegie Mellon University

Close-up of the outstretched hand of a robot

Generative Pre-trained Transformers for Crewed Uncrewed Teaming (GPT-CUT)

By Yonatan Bisk

We propose to develop algorithms to have a robot learn a new “skill” from language alone by leveraging knowledge contained in large pretrained models, both language only (LLMs) and vision language models (VLMs). Research will progress from heuristic planners and black box models to investigation of open-source LPTMs and learned differentiable planners.