Toyota conceives of more efficient method to train robots

Teaching robots entails imparting data and skills to help them execute specific jobs or adapt to different surroundings. It is a multidisciplinary field that includes a variety of techniques, from conventional programming to more sophisticated approaches like machine learning and artificial intelligence.

The most recent state-of-the-art methods for teaching robots new behaviors, however, took a lot of time, were largely ineffective and frequently restricted to specifically defined activities carried out in environments with severe constraints. Methods such as writing code that precisely specifies the robot's actions were appropriate for well defined jobs that don't call for flexibility but limited robots from undertaking more complicated tasks necessary for the functioning of machines in everyday scenarios.

The robot behavior model developed by TRI consists of learning via haptic teacher demonstrations as well as spoken descriptions of objectives. This method enables the independent introduction of new behaviors deduced from a large number of demos. The approach not only yields reliable, reproducible, and efficient outcomes, but it also does this very quickly.

Diverse dexterity

“The tasks that I’m watching these robots perform are simply amazing – even one year ago, I would not have predicted that we were close to this level of diverse dexterity,” remarked Russ Tedrake, Vice President of Robotics Research at TRI. Dr. Tedrake, who is also the Toyota Professor of Electrical Engineering and Computer Science, Aeronautics and Astronautics, and Mechanical Engineering at MIT.

“What is so exciting about this new approach is the rate and reliability with which we can add new skills. Because these skills work directly from camera images and tactile sensing, using only learned representations, they are able to perform well even on tasks that involve deformable objects, cloth, and liquids — all of which have traditionally been extremely difficult for robots.”

TRI has already used the new methodology to teach robots more than 60 intricate, dexterous abilities, without creating a single new line of code; just by giving the robots new information. TRI now hopes to teach robots 1,000 new skills by the end of 2024 and estimates that this newly-acquired ability to engage with the world in complex and varied ways will one day enable robots to support people in more unpredictable realistic contexts.

With advancements in robotics, machine learning, and artificial intelligence constantly being made, teaching robots is a dynamic field that keeps evolving. Robots are increasingly getting better at managing difficult tasks and collaborating with people in teams, making them better suited to work alongside humans in a variety of industries.