introduction Robot learning is different from machine learning. Robot learning places more emphasis on making contact with real objects in the real world, while machine learning is mostly done in a virtual environment. In machine learning, solving a robotic arm is a particularly important issue. There are 206 human skeletons and 1/4 of the hands. The hand is the most flexible part of the human body. It is necessary to create a complete intelligent robot. The smart hand is essential. This article will ask Prof. Sun Yu about the robotic arm. Your topic for today’s speech is “Rejuvenation of robot dexterous hand grabbingâ€. Can you talk about how to look at the word “revivalâ€? Robot crawling is a very old topic. The academic community did a lot in the 1980s, but it has a low tide, that is, the 90s and 2000s. I personally think that the reason is that everyone thinks they can do things. Not much, for example, I mentioned at the meeting that in 1983 the robots were already very advanced. However, there are many bottlenecks that cannot be solved because many problems cannot be solved. Many scientists went to do other things. For example, Professor Xu and other professors in the 1980s were both robotic and hand-caught, but due to problems If it cannot be solved, the research direction will be changed. In 2000 or 2005, we thought that robots or grippers were problems that had to be solved, because without solving this problem, robots became smart machines that walked on the ground or did not have arms. He could not do anything. Things can only look cool but not practical. So we just happened to have caught up with the great development of machine learning. Through machine learning to solve some of the skills of the staff, then AI has a very good development, so it is just a good time. The robotic arm of the research you studied can perceive the object material as human being and use appropriate force to pick it up. Does it use special materials or special algorithms? For this problem, there are several different directions in the field of robotics. From our point of view, we must consider the feedback of hand gripping when we grasp . The feedback of hand gripping involves a more important area called haptics. In this field, better sensors are needed. The current problem is that no cheaper sensors are used on the robot . Then you need to use machine learning to train. Your smart robot arm is a combination of machine and algorithm. Which algorithm do you prefer? For us, we think that current AI is doing well in playing chess, but the biggest problem is that AI can't have real contact with objects in nature. What we have to do is try to get the information on the contact, and then use this information to develop our own intelligent algorithm. The most important of these is to have a good understanding of the task and to express it. Your grasp must meet the requirements of the task. Your robotic arm is based on the principle of finger discoloration due to pressure caused by contact. A series of positioning, tracking, and techniques based on the shadow and depth of view, such as visual positioning. Which do you think is the most critical technique? I can extensively talk about the technology that is used here is the visual technology. This visual technology has a wide range of applications. Our philosophy is to hope that the future of visual technology can have a high-definition observation of nature or human beings. The observation we make is the change of the fingers, and the strength of a computer vision algorithm. The shadow technique you mentioned in the question is another project. The shadow is a two-dimensional camera to get three-dimensional information. We all know that when we grab something or take something, we not only need to know two-dimensional information but also know three-dimensional information, where and how high. At the time, three-dimensional cameras were very expensive. At that time, we were hoping to use shadows to better obtain three-dimensional information. Where does the color change of the fingernail cover come from? This idea was not first proposed by us. At that time we had a cooperation with MIT. What they did was a simple light sensor. We wanted to use their light sensors to do virtual reality, but found that their sensors are not very useful, because the condition light has a great influence on the fingernails, so we feel that we can use vision, so we don't need to Things can be placed anywhere on your fingernails so that you can get color changes. The human hand has 27 degrees of freedom, which is very demanding for the control algorithm. What makes the robot re-emerged? As I said today, humanoid robots still have many problems. First, the cost is high and the cost is expensive. It is not difficult to want it to re-approach, but it must solve the problem of high cost. As long as the cost comes down, other problems can be solved. If the price is low, everyone can make it, and everyone can think of ways to solve the problem. Can you tell us the difference between China, the United States and Japan in robotics? What I understand is not very comprehensive. Simply talk about my own opinion. First of all, I think that the United States has caught up with and surpassed Japan in the past two years because Japan invented many robotic things in the 1980s and 1990s, including 3D printing, and it was all Japanese patents. Now 3D printing is so fire because the Japanese patent expired. Everyone can use 3D printing, so the cost of the printer is reduced. The United States has been able to catch up with and surpass Japan because of their goals and positioning. What the United States has done during this period of time is artificial intelligence. Everyone knows that Japan's mechanical design has done a very good job in this area, but their software development is not very good and has not caught up with the tide of IT development. For the United States, they are equivalent to the start of IT, IT development needs to deal with information, which led to the development of artificial intelligence, use AI to process information. Naturally, the American AI will develop faster than other countries. I think the key to Japan lies behind the United States in terms of software and IT. I think the advantage of China is that it does not have any limitations that will be done well. What you can do with what you want to do is to pursue both. For example, some companies in the South do gear well. Baidu and other big companies do AI well. Although these two aspects have a certain gap with the United States and Japan, they have done a good job in both areas. One of your classes is for a medical school student. What is the difference compared to the classes for CS and EE students? We had some cooperation with the medical school at that time, and the professor of the medical school believed that the robot would have a great application in the medical field. Traditional medical training did not expose the students of these medical schools to new technologies. They may have the opportunity to learn about Da Vinci's medical robots. However, not all people have access to them while training in medical schools. Therefore, the professor of medical school hopes that through this course, medical students will know about these high technologies, and later they become doctors and in what areas they can make medical development better. What are the factors limiting the development of robots? The first is the issue of cost. Not all people can get expensive robotic hands to conduct research. For example, a robotic machine bought in my lab costs me 30,000 US dollars. This is still very simple with only 4 degrees of freedom. Another problem is how to control these degrees of freedom, and how to get visual results, and use the visual results on the fingers. However, both of these aspects have been greatly resolved. People have started to invent a simple robot with fewer joints but can do a lot of things. Second, there has also been great progress in the visual field, such as Microsoft's 3D camera, which makes it easier to obtain 3D information. Which direction or area is better for the robot grab? As I mentioned earlier, robotic hand grabbing is a problem that must be solved. I think the most influential area is the service robot. If the service robot has no way to grab something and hand him over to the guest, this is an incompetent service robot. What are the current technological breakthroughs and development limitations? The limitation is that I hope everyone understands that computer intelligence and robot intelligence are not exactly equal. Computer intelligence is now doing well in virtual worlds, such as games or chess, in areas that do not require physical or physical contact. IT is all about information exchange and it is dealing with data. Artificial intelligence is handled well in these areas, but once it comes to real contact, there will be many problems. Contact with them will happen in a matter of a few minutes. When you grab something, if you only have a little difference then the result will be very different. In addition, the biggest technical breakthrough I think is that robot dexterous hands can use tools. The focus of previous research is to transfer after grasping or grabbing. The reason why humans are more advanced than other animals is because we can make and use tools, rather than just picking books or catching other things. We realize that this will be followed by a better understanding of the task. I think that AI and learning are very important for the understanding of the task. Through the understanding of the expression after understanding the task, we can choose the optimal grasping program. Guangzhou Ehang Electronic Co., Ltd. , https://www.ehangmobile.com