Glossary
Foundation model
A single large model trained on a lot of data that's then adapted to many downstream tasks.
A foundation model is a large neural network trained on a broad corpus (text, images, robot trajectories, or a mix) that serves as a starting point for many downstream applications. Specialization happens via fine-tuning, prompting, or adapters rather than training a new model from scratch.
In robotics, a "robot foundation model" is the same idea applied to bodies: one large model that can be adapted to drive different robots and different tasks, instead of one model per task.