Toyota Research Institute (TRI) has developed a system that utilizes generative artificial intelligence (AI) to teach robots how to perform individual tasks necessary for preparing breakfast. Instead of relying on traditional coding methods, researchers equipped the robots with a sense of touch, connected them to an AI model, and demonstrated the tasks to them, similar to how a human would.

The sense of touch plays a crucial role in allowing the robots to “feel” what they are doing and gather more information. By providing robots with a tactile feedback mechanism, they can carry out challenging tasks more easily than if they relied solely on visual information. The TRI team refers to this technique as the creation of “Large Behavior Models” (LBMs) for the robots. Just as language models are trained by observing patterns of human writing, LBMs would learn through observation and could generalize and perform new skills that have not been explicitly taught to them.

Using this technique, researchers have already trained the robots to perform over 60 challenging skills, such as pouring liquids, using tools, and manipulating deformable objects. Their goal is to increase this number to 1,000 by the end of 2024.

Other companies, such as Google and Tesla, are also conducting similar research to teach robots to infer tasks based on their experiences. The aim is for AI-trained robots to eventually perform tasks with minimal instructions, similar to how humans can understand general directions.

However, training robots in this manner is still considered slow and laborious. Providing enough training data and ensuring accuracy can be a challenge. Nevertheless, with advances in generative AI, researchers are making significant progress in teaching complex skills to robots without the need for extensive coding and debugging.

– “Toyota Research Institute using generative AI to teach its robots how to make breakfast” – The Verge
– “Researchers Teach a Robot How to Pour From a Bag” – The New York Times