Toyota Research Institute shows off its robot AI strategy, which eliminates the need for manual coding and uses the sense of touch to efficiently learn new skills

What needs to be rewritten is: smart device

The content that needs to be rewritten is: Compilation | Chen Jiahui

The content that needs to be rewritten is: Editor | Xu Shan

What needs to be rewritten is: Smart device news on September 21, according to The Verge, Toyota Research Institute (TRI) successfully developed a breakthrough generative AI technology, a robot behavior model, in the "Robot Kindergarten" Teach robots various dexterous skills. This robot behavior model is based on AI's diffusion strategy and can learn new skills after dozens of demonstrations of new skills.

Using this robot behavior model, the robot can produce consistent, repeatable and high-performance results. In addition, the robot learns and produces very quickly, eliminating the need for manual coding and fixing coding errors, improving the robot's practicality and taking a step towards building a robot's large-scale behavior model (LBM)

Robots can also go to school and master more than 60 challenging skills by imitating robot behavior models

According to TOYOTA’s official website announcement, in the past, the technology to teach robots new behaviors was slow and inefficient, and generally had many restrictive conditions when performing tasks, and the tasks performed were also very specific and detailed. Robotics experts spend a lot of time writing complex code, possibly constantly fixing errors in the code, to add new behaviors to their robots.

Now, there is a robot behavior model. The researchers believe that touch is a key element of this robot's behavioral model. They provided the robot with a thumb-sized tactile sensor that allows the robot to sense and learn through touch, allowing it to perform various complex tasks more easily like humans. According to The Verge, researchers have taught robots how to make breakfast in a "robot kindergarten".

The way the "robot kindergarten" works is that a "teacher" first demonstrates a set of skills, then the robot's model learns in the background for several hours, and finally the robot successfully forms a new working behavior. Ben Burchfiel, manager of the Dexterous Manipulation Laboratory at Toyota Research Institute, said they often teach the robot in the afternoon, let it learn overnight, and then see its new behavior the next morning.

So far, researchers have used robot behavior models to successfully train robots to acquire more than 60 challenging skills, such as pouring liquids, using tools, and manipulating deformable objects. And they hope to increase this number to 1,000 items by the end of 2024.

▲Robot stirs liquid (Source: TOYOTA official website)

2. Independent observation and learning of new skills, a large behavioral model of the robot is being created

After developing the current robot behavior model, researchers at Toyota Research Institute are also trying to create large-scale behavior models for robots. Russ Tedrake, vice president of robotics research at Toyota Research Institute, said that large behavioral models are similar to large language models in that they learn through observation and are then able to perform new skills that have never been taught.

Google is actually researching similar technology in the development of its AI learning model robot Transformer RT-2. Similar to the Toyota researchers' approach, their robot uses its own acquired experience to reason about how to perform a task. In theory, AI-trained robots could eventually be given a general direction to complete a task without any specific instructions for action, such as "clean up that spill."

According to the New York Times, Google still has a lot of work to do in robot research and development. At the same time, the Times also pointed out that research work is often "slow and laborious", and providing enough training data is more difficult than downloading data from the Internet to train AI models

Conclusion: The robot acquires a sense of touch to increase the speed of skill learning. In the future, the robot may be able to independently develop new skills

The robot behavior model of Toyota Research Institute allows the robot to gain a sense of touch. Compared with the manual coding and error finding of training robots in the past, the new model not only reduces the labor effort, but also increases the speed at which the robot learns new skills, allowing the robot to Help people get more done, faster and better.

Researchers at Toyota Research Institute said they are building large-scale behavioral models of robots, and Google has been trying to develop similar technology. With the continuous exploration of research institutes and technology companies, large-scale behavioral models of robots may be realized in the future, allowing robots to observe and independently develop new skills

The rewritten content is: Source: The Verge

The above is the detailed content of Toyota Research Institute shows off its robot AI strategy, which eliminates the need for manual coding and uses the sense of touch to efficiently learn new skills. For more information, please follow other related articles on the PHP Chinese website!