Toyota uses generative AI to teach robots household skills

Toyota Research Institute (TRI) has applied generative AI to a "robot kindergarten" to teach them the individual tasks needed to prepare breakfast.

Here's What We Know

Instead of hundreds of hours of programming, the researchers endowed the robots with tactile sensors and connected them to an AI model. They then demonstrated the process of cooking breakfast to it.

According to the developers, touch helps the AI robot "feel" the actions and learn better.

Here's How It Works

First, the "teacher" demonstrates a set of skills, and then the model is trained for several hours in the background. Typically, researchers instruct the robot in the afternoon, leave it overnight, and check the result in the morning.

The developers said they aim to create "large behavioural models" (LBMs) for robots. Using this process, they have trained more than 60 complex skills like pouring liquids and using tools.

By the end of 2024, they want to increase that number to 1,000.

Source: Toyota