Foundation model for robotics

Instead of training a separate model for each robot and job, a company trains one big model on a lot of data from many machines, then fine-tunes it for a specific robot. The idea is the same as with large language models, applied to movement and perception.

Practical meaning: a robot that uses such a model may learn a new task from a few dozen demonstrations rather than months of engineering. Check whether the model is open, whether you can fine-tune it on your own tasks, and what the vendor charges for it.