Nowadays, the combination of biomimicry and application needs in the field of robotics is an advanced stage and an inevitable trend in the development of robots. Bionic multi-legged robots such as humanoid robots, quadruped robots, and spider robots have good terrain adaptability, strong environmental adaptability and ground motion stability, and interdisciplinary technical myths have attracted a group of academic colleagues to pay attention to this field.
Similarly, the dynamics of bionic robots in the industry have also made bionic robots gradually become the focus of the public, especially after elon Musk released the "Tesla humanoid robot", Xiaomi released the four-legged robot "Iron Egg" and Xiaopeng released a rideable quadruped machine horse, companies with various industry backgrounds participated in it, making the biomimetic ecology of the robot more sound. But at present, whether it is body perception and body control, bionic robot technology is still in the stage of "looking beautiful", and more people need to participate in this industry-academia discussion.
In the industry-university linkage, CNCC has set up a new bionic robot forum this year, which will be held at CC105C of Shenzhen World Convention and Exhibition Center from 13:00 to 16:00 on October 28. The forum was co-chaired by Dr. Pang Jianxin, Vice President of Shenzhen UBTECH Technology Co., Ltd. and Vice President of Shenzhen Research Institute, and Dr. Huang Weiwei of Shenzhen Artificial Intelligence and Robotics Research Institute, and invited famous scholars from scientific research institutions such as Shenzhen Institute of Advanced Technology, Southern University of Science and Technology, Hong Kong university of Chinese (Shenzhen) of the Chinese Academy of Sciences, as well as industry experts from high-tech enterprises such as Tencent and UBTECH to conduct in-depth analysis and share the development status of a variety of bionic robots. The application prospect of the joint pulse bionic robot.
Before the forum began, Lei Feng Network & AI Technology Review invited two forum chairmen, Pang Jianxin and Huang Weiwei, to discuss the starting point of CNCC's new bionic robot forum this year, as well as the technical difficulties and industrial opportunities of advanced robots such as bionic robots.
The following is a transcript of the interview:
<h3>First, what are the preparations for this year's bionic robot forum? </h3>
AI Technology Review: The two teachers served as the chair of the CNC Conference Bionic Robot Forum this year, can you introduce the specific content of the organizational work you are involved in?
Huang Weiwei: The conference was mainly led by Mr. Pang, because UBTECH has been focusing on the research and development, manufacturing and sales of artificial intelligence and humanoid robots for nearly a decade, and has long received attention in the field of bionic robots. My Shenzhen Institute of Artificial Intelligence and Robotics (AIRS) also has a lot of research on bionic robots, so the two of us have jointly launched such a forum around bionic robots.
This year's bionic robot is also quite popular, in the early stage of preparation, I and Mr. Pang together selected a guest candidate, hoping to invite industry experts, scholars and entrepreneurs in the field of bionic robots to attend the meeting, through the forum to stimulate the collision of ideas, together to discuss the future trend of bionic robot technology development and industrial application.
AI Technology Review: What considerations do the two teachers think the new bionic robot forum at this year's CNCC conference is, is it because there is more research progress in this regard, or is it trying to unite more industrial opportunities?
Huang Weiwei: You may wonder how the computer conference can set up a robot forum, in fact, computers and robots are two closely linked parts. CNCC itself is a relatively open computer conference, compared with previous years, this year's conference set up 113 forums, reaching the highest in history. At this time, we launched the bionic robot forum, one is to pay attention to the forum in previous years has not been involved, and this is a new direction of common concern between industry and academia. Strictly speaking, at present, bionic robots are more popular in the industry, but the academic research circle has actually paid attention to this field a long time ago, and now organizing everyone to discuss together is actually a good opportunity to seize the public attention. Second, the bionic robot is an ideal carrier of "intelligence", which is closely related to computing.
Pang Jianxin: Now everyone can also see that various disciplines are converging towards multidisciplinary development, such as robots, artificial intelligence, and computers, which are breaking down the structural barriers of disciplines. Bionic robot is a representative complex system of multidisciplinary integration, so we hope to set up a forum for bionic robots to share the progress of this exploration of multidisciplinary integration with you at the meeting.
At the same time, we also hope that through this multidisciplinary cross-integration to form a collision, it can help the multidisciplinary of bionic robots, and attract more people engaged in computer science and other disciplines to participate in this field. Whether it is for academic research or industrial incubation, it is a role that the forum wants to play.
AI Technology Review: Teacher, you just said that you hope that more people will participate, so what is the biggest feature of this year's humanoid robot forum, and what are the plans for guest invitations and report topics?
Huang Weiwei: As Mr. Pang said, the bionic robot itself involves the integration of multiple disciplines, including AI, computing, robotics, etc., so when we plan the forum, we also hope that the guests can have various professional backgrounds, such as control, institutional design, materials and so on. Professor Wu Xinyu in the Shenzhen Institute of Advanced Technology of the Chinese Academy of Sciences and Professor Zhang Wei of the Southern University of Science and Technology are not only at the forefront of academia, but also have very prominent research results in the most critical field of robots - control.
At the industry guests, not only Mr. Pang of UBTECH Technology, but also Invited Zheng Yu, a researcher at Tencent Robotics X, who will introduce some of the lab's explorations in the design of foot robots, such as the multimodal quadruped robot Max and the roller skating robot Ollie. Leg-foot robot is an important branch of bionic robots, and Dr. Luo Jianwen, associate researcher of the Institute of Robotics and Intelligent Manufacturing of hong Kong Chinese University (Shenzhen), will sort out the development trajectory, engineering challenges and theoretical development of leg-legged robots at home and abroad in the past 20 years at the forum.
In addition, we thought of other more interesting bionic forms, such as soft bionics, so we invited Professor Zhu Jian, who is an associate professor at the School of Technology of Chinese University of Hong Kong, Shenzhen, who has long been interested in soft intelligent robots, bionic robots, and intelligent materials and structures. He will focus on the preliminary work on soft actuators and soft robots in recent years, including soft underwater robots and soft crawling robots, and will also introduce the recent research progress of humanoid robots.
Pang Jianxin: In fact, one of the core starting points of bionic robots is how to simulate organisms. There are human-machine combination robots that imitate humans, such as exoskeleton robots; there are humanoid robots, bipedal robots, wheeled foot hybrid robots; and soft robots that imitate mollusks, etc., which are essentially imitating organisms from different dimensions.
Of course, foot robots may be more abundant, because these two years of bipedal humanoid robots and multi-legged robots are particularly popular, UBTECH's large bipedal human service robot Walker has been 5 years old, in July this year in WAIC released a new generation of Walker X, in September released for the Dubai World Expo China Pavilion specially customized noh cat robot Youyou, this year's Year of the Ox Spring Festival Gala also showed the quadruped robot pioneer cattle for the first time, other companies such as Xiaomi released a four-legged robot dog, Tesla also predicted to do humanoid robots and so on, so we put the foot robot in a particularly important position in the forum, which actually matches the current research heat, especially the research heat of the industry.
<h3>Second, where is the research and development and application scenario of humanoid robots? </h3>
AI Technology Review: Teacher Pang, the report on your forum specifically talked about the technical practice and application exploration of humanoid robots, you observed that the current domestic and international humanoid robot research and development and industry, how mature are they, and how big is the gap between domestic and foreign countries?
Pang Jianxin: My personal understanding is that in fact, the discipline of humanoid robots was first studied from abroad, and in recent years, domestic foot robots have begun to really exert their strength. The earliest is the bipedal robot of Beijing Institute of Technology, which should be one of the earliest bipedal humanoid robots in China. Later, the national university of science and technology, Harbin Institute of Technology, Zhejiang University, Hong Kong Chinese and other school laboratories and some scientific research institutes began to engage in the research of bionic robots.
Among domestic enterprises, UBTECH has gone relatively early, and the bipedal humanoid robot is a particularly representative product of ours. After that, there are also many startups that invest in research on small humanoid robots and large humanoid robots, which is some development progress in China.
The progress of overseas humanoid robot products is mainly driven by European and American companies, the earliest commercialization is more successful is France's Aldebaran Robotics, which was later acquired by Japan's SoftBank, they are a successful model of small bipedal robots, launched the Nao robot. There are also some more famous companies such as Boston Dynamics Atlas, Honda's ASIMO, South Korea's Hubo, and the Cassie robot launched by Agility Robotics in the United States, which are very representative works in the field of bipedal robots.
My understanding is that the maturity of the humanoid robot should be seen from two dimensions, whether it is a foot-type humanoid or a non-foot-type humanoid. If it is a non-foot imitation, its product form is relatively mature, because it can be designed as a bionic system with lower body wheels and upper body humanoids, which actually has many commercial scenarios.
The foot-type humanoid robot is still in the stage of scientific research and exploration, and it is also doing some commercial exploration. It can be divided into small foot humanoid robots and adult-sized foot humanoid robots, and this small one has been used in the education market and the toy market, and the technology and market are relatively more mature.
I think that the current marketization of small foot-type humanoid robots is better at home than overseas. The large-scale, adult-sized humanoid robots, foreign technology has gone earlier, with certain advantages, but the commercialization road is not smooth, but we can see that the country is also catching up. However, in general, as a complex system integrating various disciplines and technologies, humanoid robots at home and abroad still need to be upgraded in an all-round way, and need to be promoted in the fields of materials, control, mechatronics, energy, AND. In particular, the product of the robot, its product experience and user experience, is actually determined by its worst part, and it is expected that the technology of various disciplines will make great progress.
AI technology review: Before Teacher Huang did a service robot in this area of smart property, UBTECH technology this year also released a new generation of large-scale humanoid service robot Walker X and the panda robot Youyou specially customized for the World Expo China Pavilion These bipedal robots are unanimously optimistic about service robots. For humanoid robots, why is it very suitable for service scenarios, and UBTECH will focus on family scenarios?
Pang Jianxin: That's how I look at this problem, and I believe that this kind of humanoid service robot is most suitable for use in family scenes.
We humans have evolved through millions of years of development, and have continuously evolved into the current human form through biological evolution, which determines that our current form must have some innate advantages, and I understand that this form has the best environmental adaptation ability. There is no doubt that we can't run better than leopards, leopards can't run cars; we can't swim better than fish, and fish can't swim boats, but in fact, the reason why we humans are still able to dominate the current world is because we have the best environmental adaptability. Similarly, humanoid robots are able to express human form and movement ability, adapting to a variety of man-made scenes and natural scenes.
People have two appeals to robots, one is a psychological appeal. For example, people naturally have a sense of closeness to humanoid robots, and it is easier to produce a sense of companionship or psychological satisfaction. The second is the appeal of the transaction, which does not mean that we need a robot to help wash clothes, there is no doubt that the washing machine washes better than the "robot". But humanoid robots can help us solve some simple, repetitive home tasks and tasks in family scenarios, such as managing home devices; when you eat and watch TV, robots can also intelligently and intimately help you get a glass of water.
Of course, we can also see that due to the limitations of various technologies, humanoid robots are not yet very robust to adapt to this scene. The purpose of our humanoid robot, in addition to hoping that this scenario can be landed in the future, we also hope that it will serve as a major technology research and development route of the company, driving the incubation of a series of technologies, such as control technology, mechatronics technology, and AI technology. Then, relying on all these technologies, a series of robot products for home or commercial scenarios are developed. For example, in addition to the service robot of Walker, UBTECH also has a series of service robots based on practical application scenarios such as health care, education, logistics, and epidemic prevention.
The problem of the scene ultimately comes back to the technical problem. Because these technologies for humanoid robots are not the same as other fields of technology, such as vision technology, positioning and navigation technology, etc., and various technologies must also be integrated. Therefore, we will develop our own AI technology for humanoid service robots, so that we can better couple with our robot body and smoothly perform tasks in various scenarios.
Huang Weiwei: I think that whether the humanoid robot is a companion to people or a coexistence with people, the key question is its form, whether it is a more "machine" form, or a more "human" form. Because the machine does not have the attribute of companionship, but the humanoid or bionic robot, because it has a more "biological" attribute, so companionship has become a "feature" after its manufacture.
AI Technology Review: There was a saying before that if a humanoid robot appears in a family scene, humans may have a kind of "empathy" for it, and UBTECH is now launching these service robots, is it possible to transform this concern into an advantage?
Pang Jianxin: When AI technology or robot ontology technology does not reach or exceed the human level, empathy problems between humans and robots are unlikely. At the same time, I think that we do robot research and development, products or good, must have moral constraints and laws and regulations, the current corresponding regulations are relatively missing, but I believe that with the progress of technology, it will have a formulation process. Before the technology and regulations come out, the first thing is to abide by the duty and principle of scientific and technological workers, and adhere to one: to use technology to serve people, rather than using technology to kidnap people.
AI Technology Review: Combined with the actual situation of the current bionic robot, where will the application scenario of the simulation robot be in the future?
Pang Jianxin: There are several stages, the first stage is such as the panda robot Youyou, which is used for the display and reception of high-end exhibition hall scenes, which is a scene that can land in the short term. As the "messenger of peace and friendship" of the Chinese Pavilion of the Dubai World Expo, Panda Robot Youyou is currently sought after by audiences from all over the world in Dubai. In the medium term, it may be applied to high-end family scenes, and even slowly gradually enter the commercial scene, and finally slowly enter the daily family scene. When the cost or price of a robot is about the same as the price of a medium car, robots can begin to become more popular.
AI Technology Review: For UBTECH, what is the technology planning and product layout for the next year and five years?
Pang Jianxin: Our robot products should be based on our technical planning at different stages, and the core capabilities to be solved at this stage are divided into three parts. The first is the robot body, how to build a more efficient or more stable body, including mechanical, electromechanical structure, servo drive and the topology of the entire robot. The second block is to build the robot's movement ability, that is, the robot's motion planning and control technology, how to move steadily in the scene. The third block is AI technology for robots, especially perception and interaction capabilities, mobile manipulation capabilities, etc.
<h3>Third, what is the most critical technology of bionic robots? </h3>
AI Technology Review: Teacher Pang, you mentioned a number of artificial intelligence technologies such as vision, navigation, speech, control, drive, and structural design in the abstract of the forum' report, which branch currently has the most problems and what problems need to be solved?
Pang Jianxin: In fact, it is difficult to say which technology is the most difficult, because each technology involves computing and some of the technologies related to it. These branches are not exactly separate branches. For example, the positioning and navigation technology of the simulated robot needs to integrate control and perception technology.
Voice interaction technology involves computing, but due to the relatively high requirements for computing power, it is necessary to consider putting part of the computing in the cloud, and part of the computing to the edge side, or the end side. And because computation involves a large number of different types of knowledge and data, there will be the need for heterogeneous computing. At the same time, there are also multi-modal interactions, emotional perception and other needs at the voice level, so each branch of technology is a very challenging task, especially in the field of robotics.
AI Technology Review: Some people mentioned that in the future, artificial muscles and artificial skin will gradually appear, and physical learning and tactile - perception ability will make robots go to the next level. The recent Nobel Prize found that the connection between vision and proprioception requires a molecular-sized, gate-like receptor (piezo2), through binocular vision, and through this receptor, the physical force can enter the nervous system, convert into electrical impulses and then rise to consciousness. The two statements mention two perceptual pathways, skin perception and visual perception, what do you think is the best perceptual path for the future development of bionic robots?
Pang Jianxin: There is no doubt that humanoid robots need multi-dimensional perception and establish a three-in-one perception system of vision, hearing and touch.
If it is visual perception, the easiest way is to install more visual perception devices on the robot. People are binocular perception, can only perceive the front, and there is a blind spot of focus. Then the humanoid robot can establish a 360-degree perception ability through poly-eye vision technology; infrared perception technology can establish a perception technology for invisible light.
The second is the auditory perception system. Humans can distinguish the location and content of occurrence through ambient sounds and interpersonal interaction sounds, but the current source of auditory perception of robots comes more from interaction with people.
The third is haptic perception, when the robot encounters an object, it can sense the force through the limb, and then output the force through the joint to complete the control task. Because when the robot performs the task, there must be a visual blind spot or a non-structural human environment, so at this time, the robot can have tactile perception, which has very good application value.
AI Technology Review: Just mentioned that these artificial intelligence technologies for robots are actually very important, and Teacher Huang is currently focusing on AI research. What are some of the priorities of the Shenzhen Institute of Artificial Intelligence and Robotics, where you are now located, and what are the research progress?
Huang Weiwei: The Shenzhen Institute of Artificial Intelligence and Robotics, where I am now, is one of the top ten basic research institutions in Shenzhen, and in the field of bionics, the research institute team has done some very good work, including climbing robots based on the crawling mechanism of caterpillars, which can be used for bridge detection. We are also making haptic sensing materials now, and we are already able to have three-dimensional force perception capabilities. We also explore federated learning between multiple robots, because what a robot can learn is after all a minority, and if there are thousands of robots distributed in different scenarios, each robot learns a single function, and then uses federated learning to aggregate skills, then they can evolve rapidly.
I am engaged in the research of the intelligence of robots, with a focus on the field of AI. I originally turned out to be a humanoid robot and the control of the robot, mainly based on the dynamic control of the robot's whole body joints. This is also a difficult subject, but robots are a complex system, and one of the shortcomings is the intelligence part, so I moved from robot control to research on robot intelligence. Of course, these two parts can also be well combined, such as reinforcement learning and large-scale pre-training models. Large-scale pre-trained model research has achieved remarkable results in language processing and vision, and it is not known whether it can be combined with complex systemic joint dynamics control to form a new breakthrough in robot control.
AI Technology Review: Teacher Huang, you said that what you are mainly doing now is research in the field of AI, involving large-scale pre-trained models, rule learning, federated learning, etc. Now you see, how does AI empower bionic robots?
Huang Weiwei: For robots, one is the motor function of the body, and the other is intelligence. At present, a big constraint for robots is adaptability. Humans and animals are adaptive, and when we encounter some situations, we will adjust in time and in real time according to the feedback of the environment. So how to achieve adaptive control requires learning to refer to the characteristics of neural responses and motor functions of humans and animals. An important aspect of this adaptive ability is perception. Now the most widely used is based on deep learning visual perception, visual algorithms in recognition, segmentation and posture estimation and other aspects of good accuracy. The sensing of the skin plays a very important role in the control of organisms. For a large-scale and relatively complex system, if it has many signal sources and it is more difficult to do quantitative analysis, AI's deep learning model can help the group simulate skin perception.
Another aspect is decision-making, where we use AI to better simulate decision-making to achieve adaptive. At present, our research on the brain is still in a very unclear process. Whether it is the motion mechanism of the human brain or the information processing mechanism, many times it can only be simulated approximately through the results we see. In the future, if there is a breakthrough in robot-oriented technology, it is necessary to have a theoretical breakthrough or biological reference, so that we can use AI to simulate more directional.
AI technology review: Before participating in many competitions have obtained good results, like the AMERICAN DARPA competition, the study of humanoid robots has been more than ten years, what do you think is the most critical technical difficulty of bionic robots, what is the evolution trend?
Huang Weiwei: Now bipedal robots are more challenging, after all, its dynamics are more complex. The quadruped is simpler, because it does not need to think too much about dynamic stability, as long as there are three feet landing, the stability and support of the entire robot is very good. This is also why there are more quadruped products on the market.
I think the difficulty of the bionic robot may still be the original mechanism, because the current motion control, whether it is motor control or the hydraulic control used by Boston Dynamics, is actually wanted to get "muscle control". But at present, there is no good bionic robot product to achieve force control. Like motor control, it is much different from biological control in terms of rigid-soft coupling.
In short, if the bionic robot has new achievements in artificial muscle, joint control, multimodal perception, and tactile perception, it will promote the development of the entire bionic robot.
<h3>Fourth, the conference is coming, and industry and academia are talking together</h3>
Has the spring of bionic robots arrived? What is the technological innovation of bionic robots? What kind of new ideas will experts and scholars collide? Under the background of the conference theme of "Computing Empowerment Accelerates Digital Transformation", what on-site aura will be stimulated by the seven big coffees from the fields of academics, technology, industry, education and other fields in the Bionic Robot Forum?
At the setting of this year's bionic robot forum, seven teachers from [Wu Xinyu, Zhang Wei, Pang Jianxin, Huang Weiwei, Zheng Yu, Zhu Jian, Luo Jianwen] made sharing reports respectively, and then set up a roundtable forum to discuss the application prospects and development of bionic robots. We look forward to the collision of ideas brought about by this forum, and at the same time, we are also looking forward to the new industry-university interaction that the bionic robot will usher in after the teachers talk together!
CNCC2021 will be held in Shenzhen on October 28-30, and the theme of this year's conference is "Computing Enabling To Accelerate Digital Transformation". CNCC is an annual event in the field of computing academia, industry, and education, which explores the development trend of technology macroscopically, and the number of participants is expected to reach 10,000 this year. The guests of the annual specially invited report bring together academicians, Turing Award winners, scholars from famous universities at home and abroad, leaders of famous enterprises, and influential industry experts in various fields, and the luxurious guest lineup highlights the top industry level and industry influence of CNCC.
This year's special guests include ACM Turing Award winners Professor John Hopcroft and Professor Barbara Liskov, Professor Yolanda Gil from the Department of Computer Science and Space Research at the University of Southern California, Academicians Chen Weijiang, Feng Dengguo, Guo Guangcan, Sun Ninghui, Wang Huaimin, and many influential experts in the industry. This year's technology forums as many as 113, whether in terms of quantity, quality or coverage, have created the highest in history, will bring participants a full range of academic, technical, industrial, educational, popular science and other aspects of the experience. During the conference, a large-scale theme carnival of "Members' Night" will be held for the first time to allow participants to communicate freely.
CNCC2021 will bring together top professional forces and expert resources at home and abroad to present a wonderful and grand professional feast for more than 10,000 participants!
Lei Feng network