Toyota Research Institute (TRI) used generative artificial intelligence to teach robots how to cook breakfast. This did not require hundreds of hours of coding and fixing errors, but simply connected the robots to an artificial intelligence model and trained them using it.
According to the researchers, the aspect of added touch sensitivity was also important in this project. If the model can “feel” what it is doing, it gets more information, and this makes difficult tasks easier than by sight alone.
Researchers are trying to create large behavioral models, or LBMs, for robots. Similar to how LLMs are trained by observing people’s writing patterns, Toyota LBMs learn by observation, then generalize by acquiring a new skill they’ve never been taught.
Using this process, researchers say they have trained more than 60 challenging skills, such as pouring liquids, using tools, and manipulating deformable objects. They want to increase this number to a thousand by the end of 2024.
Source: The Verge