Español
Português
Việt Nam
日本語
русский 
عربي 
Deutsch
한국어
Türkçe
Français
Italiano
AI+ robots are moving towards reality step by step.
Not long ago, Figure humanoid robot demonstrated the autonomous operation of the coffee machine, only need to learn end-to-end 10 hours, the robot can accurately operate the coffee machine, without any remote control, and has the ability to automatically correct the error, can recognize the unplaced coffee bag, and put the coffee bag itself.
Not to be outdone, Toyota, the world's second largest car company, uses AI technology and machine learning systems to enable robots to observe and learn human behavior in housework activities, and imitate humans to do housework. This project not only demonstrates the progress of generative artificial intelligence to help robotics, but also provides new solutions for the future of smart homes and elderly life.
From the public video of the experiment, the experimenter, from the perspective of a person who likes to organize, controlled the remote control with a mechanical arm, so that the robot learned how to use a dustpan and brush to sweep away the beans on the table, and the robot's own sensors allowed the robot to sense the grasp of solid objects and the weight of objects. After teaching, you can complete this housework activity independently.
On this robotic arm, Toyota employs a machine learning system called diffusion strategy, which is similar to the system used in artificial intelligence image generators, to quickly suggest the right next action through multiple possibilities and information.
The technology, developed in collaboration with researchers led by Columbia University professor Shuran Song, offers robots a more efficient way to learn.
The use of such machine learning systems allows robots to absorb data more flexibly, thus better adapting to the complex physical world and human environment, and makes robots better suited to work requiring adaptability, improvisation and flexibility than traditional pre-programmed routines.
Toyota is trying to combine machine learning systems with language models that support ChatGPT and its competitors, by combining language models that robots can learn to perform tasks by watching videos, turning online resources such as YouTube into powerful robot training resources that can enable robots to learn more tasks by watching videos. So as to improve the efficiency of learning.
This is not the first robot to learn skills through YouTube, a team of researchers in the UK has launched a similar robot chef, the team implanted a new learning algorithm for this robot chef, based on the visual observation of the human chef, gradually add the recipe to the robot's recipe, by calculating the similarity between the vectorization of the two. Learn and add new recipes in real time.
In subsequent experiments, the team fed the robot eight simple salad recipes and had it watch videos of humans preparing the salad. After watching these videos, the robot was able to recognize the recipe used and replicate it.
The robot analyzed every frame of the video, identifying various objects such as a knife or the presenter's hand, an ability that allowed it to infer what would happen next - for example, if the presenter was holding a knife and a carrot, the robot could predict that the carrot was about to be chopped up. After the robot watched 16 videos, the algorithm correctly identified known recipes 93 percent of the time.
It is worth mentioning that the team that developed the robot is from the Cambridge Robotics Institute established by Toyota in 2015.
Applying language models to robotics research seems to be a new trend these days, with some big tech companies, such as Google DeepMind, pushing for similar research. A team at Google DeepMind has released Auto-R software, which uses large language models to help robots determine practical and safe tasks, such as simply sorting items and picking up a water glass.
Taking the express train of generative artificial intelligence, robots have been gradually realized to help humans share housework activities through learning, which will bring more convenience to our lives, and also promote the commercialization of service robots such as hot humanoid robots. The future may be true as Musk said, each family will have several robots working for us, and these smarter and more flexible robots will become important partners in our lives.