The technology that Toyota has created, capable of teaching hundreds of new skills to robots, efficiently and in a very short time.
All the robots that we see on assembly lines, or even in places like restaurants, have had to go through a prior training process in neural networks that has given them some progress in their tasks.
However, this pre-training process can even take years, but luckily, Toyota Research Institute (TRI) has created a generative artificial intelligence method, which allows you to quickly and efficiently include new and improved skills to robots.
“Our robotics research aims to amplify people rather than replace them,” he said. Gill PrattCEO of TRI and Chief Scientist of Toyota Motor Corporation.
“This new teaching technique is very efficient and produces very high-performance behaviors, allowing robots to amplify people in many ways more effectively.”
The field of teaching robots includes a wide variety of techniques, ranging from conventional programming to more sophisticated approaches such as machine learning.
The bad thing is that the most advanced methods of teaching robots require a lot of time, and some of them even prove to be ineffective.
This is how this new technology works
This new technology created by Toyota is learning through haptic demonstrations, as well as spoken descriptions of the objectives.
This allows new behaviors to be introduced independently to the robots, producing reliable, efficient, and fast results.
“The tasks I’m seeing these robots perform are simply amazing; Even a year ago, I wouldn’t have predicted we would be anywhere near this level of diverse skill,” he says. Russ Tedrakevice president of Robotics Research at TRI.
“What’s interesting about this new approach is the speed and reliability with which we can add new skills. Because these abilities work directly from images from cameras and touch sensors, using only learned representations, they can perform well even on tasks involving deformable objects, fabrics, and liquids, all of which have traditionally been extremely difficult for robots,” Add.
So far they have used this new methodology to teach their robots more than 60 complex skills, and the best of all is that they have not needed to create a single new line of code.
Now they hope that their technology be able to teach robots 1,000 new skills by the end of 2024, and they believe these skills will allow them to interact with humans in complex and varied ways.