In the Toyota Research Lab laboratory in Cambridge, Massachusetts, developers > The creation of robots that are capable of independent training and performing complex tasks. One of these experiments is the use of robotic hands with bicapaltsery forking for cleaning scattered beans on the table. This process, although it seems simple, requires significant adaptation and flexibility from the robot.
The Toyota project is not limited to cleaning alone. In the laboratory, robots master various homework: from cleaning vegetables to cooking pancakes. Particular attention is paid to the use of language models like ChatGPT for teaching robots. Similar models contain information about the physical world, which helps robots understand objects and methods of their use.
The main breakthrough in the field of robot training is associated using machine learning, in particular, a diffusion policy – a method similar to those used in AI images. This method has been developed by Toyota together with researchers from Colombian University and Stanford University. It allows the robot to choose the right actions from many options in a matter of seconds, based on multiple sources of data.
Another innovation is an attempt to teach robots using video from YouTube. This approach can turn a video platform into a powerful resource for training robots. However, as the Russa Tedreik notes, vice president of robotics research at Toyota Research Institute and professor of the Massachusetts Institute of Technology, for a complete understanding of the physical world, robots still need the basic experience of interacting with the real world.
Toyota announced the creation of her institute of robotics in Cambridge in 2015, as well as the second institute and headquarters in Palo Alto, California. The company hopes that robots will help people to maintain independence with age, not only in Japan, but also in other developed countries.
Despite the impressive demonstrations and significant progress in the field of robotics, it is worth noting that robots still make mistakes and sometimes behave strangely, like the early versions of the models standing for Chatgpt.
Toyota is not the only company striving to use language models to promote research in the field of robotics. For example, the Google DeepMind team recently introduced Auto-R software, which uses a large language model to help robots in determining the tasks that they can perform in the real world safely and realistically.