Shulou(Shulou.com)11/24 Report--

On August 17, robots are now everywhere, and they are also found in factories around the world. However, these robots can only follow specific instructions and focus on simple pre-programming tasks. But Google's robots seem to be different. They can take simple commands and perform more complex tasks, and look more like humans.

Researchers at Google Labs recently demonstrated new robotic skills, making hamburgers from a variety of plastic toy ingredients. The robot understands the cooking process and knows to add ketchup after the meat and before the lettuce, but it thinks the right way is to put the whole bottle in the hamburger. While the robot won't be a smart cook any time soon, it represents a major breakthrough declared by Google.

Using recently developed artificial intelligence (AI) software known as large language models, they have been able to design robots that can help humans perform a wider range of daily tasks, researchers say. Instead of receiving a series of decomposition instructions and then instructing them to complete each action one by one, such robots can respond to complete commands, more like humans.

In last week's demonstration, Google researchers told the robot: "I'm hungry. Can you get me some snacks?" Then the robot begins to search the cafeteria. It opens the drawer, finds potato chips, and brings them to humans. Google executives and researchers say this is the first time a language model has been integrated into a robot. Brian Brian Ichter, a Google research scientist, said: "fundamentally, this is a very different model for training robots."

Currently, robots on the market usually focus on one or two tasks, such as moving a product on an assembly line or welding two pieces of metal together. It is much more complicated to develop robots that can perform a series of daily tasks and learn independently at work. For years, technology companies, large and small, have been trying to build such general-purpose robots.

The language model works by taking a large number of texts uploaded to the Internet and using them to train AI software to guess how they might react after certain questions or comments. These models have become so good at predicting correct responses that when dealing with people, they often feel like they are talking to a knowledgeable person. Google and other companies, including OpenAI and Microsoft, have invested a lot of resources in building better language models and training them in multilingual and increasing text.

However, there is a lot of controversy about this work. In July, Google fired an employee because he claimed that Google robots had human perception. The consensus among AI experts is that these models are not perceptual and many worry that they will show bias because they are trained in a large amount of unfiltered, human-generated text.

Some language models show racist or sexist tendencies, or are easily manipulated to express hate speech or lies. Dipak Patak, an assistant professor at Carnegie Mellon University, said that in general, language models allow robots to understand more advanced planning steps, but do not provide robots with all the information they need, such as how much force they need to open the refrigerator. This knowledge must have come from somewhere else. "the language model only solves the problem of high-level planning," Patak said. "

Still, Google is moving forward and has integrated its language model with many of its robots. Now, researchers can simply talk to the robot in everyday language, rather than writing specific technical instructions for every task the robot can perform. More importantly, the new software helps robots parse complex multi-step instructions themselves. Today, robots can interpret instructions they have never heard before, respond meaningfully and take action.

Robots that can use language models could change the way manufacturing and distribution facilities operate, says Zac Stewart Rogers, an assistant professor of supply chain management at Colorado State University. "Human and robot cooperation can always greatly improve efficiency," he said. Robots can lift heavy objects manually, and humans can troubleshoot nuances. "

If robots can handle complex tasks, this may mean that distribution centers may become smaller, requiring fewer people and more robots. It may also mean that people have fewer job opportunities. Although Rogers points out that usually, when one area shrinks because of automation, more jobs are created in other areas.

Training general-purpose robots may still have a long way to go. For many years, AI techniques such as neural network and reinforcement learning have been used to train robots. Although many breakthroughs have been made, progress is still slow, and Google Robot is far from ready to serve the real world. Google researchers and executives have repeatedly said they are just running a research lab and have no plans to commercialize the technology.

But it is clear that Google and other big technology companies have a strong interest in robotics. Amazon, which uses many robots in its warehouses and is experimenting with drone delivery, agreed earlier this month to buy the maker of the Roomba vacuum cleaner robot for $1.7 billion. Tesla says it is building friendly robots that can perform daily tasks and do not fight back. In addition, Tesla has developed a number of self-driving functions for his car, and the company is also developing a general-purpose robot.

In 2013, Google began investing heavily, acquiring several robotics companies, including Boston Dynamics, which often sparked a heated debate on social media. But the executive in charge of the project was accused of sexual misconduct and left the company shortly afterwards. In 2017, Google sold Boston Dynamics to Japanese technology giant Softbank Corp.. Since then, the hype surrounding increasingly intelligent robots designed by the most powerful technology companies has cooled.

In the language model project, Google researchers worked with the Everyday Robots team. Everyday Robots is a Google-owned but independent company that specializes in building robots that can perform a series of "repetitive and tedious" tasks. The robots already work in Google's cafeterias, cleaning counters and garbage.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.