Artificial Intelligence + Robots: New Opportunities to Improve Manufacturing Efficiency

  Entering 2019, China's AI investment and startup companies are continuing to increase. There are too many investment and financing cases of "artificial intelligence + manufacturing".

    "AI" can be seen everywhere, and has become a popular buzzword for many people. Michael I. Jordan, a master of machine learning, thinks that this phenomenon makes him very uneasy: “AI is just a concept they use to sell their own concepts to VCs, companies, the media, and the public. As for real AI, we basically It hasn't been realized yet.

    Once, in the industrial field pursuing cost-effectiveness and practicality, "artificial intelligence is only a small supporting role on the stage of intelligent manufacturing." Nowadays, regarding specific application scenarios, people in the industry generally believe that artificial intelligence will greatly improve the work efficiency of industrial robots.

    Up to now, what artificial intelligence has been found in the "new continent" of industrial application scenarios in the robotics industry? Where is the future development direction of the combination of artificial intelligence technology and robotics?

    Artificial intelligence + traditional industrial robot = intelligent robot

    Traditional industrial robots are a high degree of integration of mechanical design and manufacturing technology, automatic control technology, and computer software and hardware technology.

    Artificial intelligence is a collection of data and algorithms. The continuous increase in computing power (chips) is the basis for the widespread application of artificial intelligence. At present, artificial intelligence is still in the stage of weak artificial intelligence, and the field of breakthroughs is still relatively limited. The combination of artificial intelligence technology and robotics technology to realize a robot with both the limbs of a robot and human-like intelligence is the ultimate goal of the development of artificial intelligence and robotics. Intelligent robots are the result of the integration and development of artificial intelligence technology and traditional industrial robot technology.

    Zheng Yong, CEO of Geek+, said that if artificial intelligence is defined to the level of "deep learning", there are currently few applications. He believes that current artificial intelligence can be defined as "autonomous capabilities brought about by relatively complex algorithms."

    Geek+, which focuses on the field of robotic intelligent logistics, empowers the logistics and warehousing industry through artificial intelligence and robotics technology, through the optimization of warehousing and logistics links such as intelligent picking, handling, and sorting, and highly flexible human-computer interaction. The purpose of improving warehouse efficiency, reducing labor costs and labor intensity.

    Cooper's CEO Li Miao pointed out that "sorting, polishing, assembling, and testing" are the four most urgent and extensive areas of artificial intelligence and robot landing applications. As a result, Cooper's self-developed system can be applied to the disorderly sorting of loading and unloading, the force control polishing of mobile phones or aviation blades, intelligent teaching, intelligent labeling, and parts assembly through core learning algorithms and special control software. .

    "In the AI ??era, industrial robots will be defined by new core technologies, including deep learning, path planning, task-level programming, flexible control, etc." said Shao Tianlan, CEO of Mecamand. In his opinion, the sorting of mixed objects is currently the most obvious part of demand and the most direct application. Many companies can show a certain degree of demo, but the products that can be used on a large scale have not yet appeared.

    In addition, there is another combination point for "operation planning", that is, people only need to specify the installation requirements of multiple workpieces, and the robot can calculate the grasping and installation plan by itself, saving a lot of programming time.

    In the standard scenario, the products produced by industrial robots are large in batches, there are a lot of repetitive tasks, and high-frequency trajectory optimization is required, such as machine tool processing, parts installation and other applications. At this time, small samples can be used to supervise learning, so that the robot has adaptive and evolutionary functions.

    Earlier, Elit showed a "robot folding clothes" demo, showing that robot trajectory optimization can not only target rigid objects, but also deal with flexible objects such as clothes. Elite’s robotic clothing stacking system uses deep reinforcement learning algorithms and deep vision sensors to precisely locate the clothing stacking points, automatically optimizing the best motion trajectory, and achieving stacking effects. The system also uses a simulation environment for rapid modeling and migration learning methods to speed up learning, reduce data collection costs, and finally map the simulation results to real robot operations.

    In addition to the above-mentioned applications that focus on improving the efficiency of industrial robots, machine vision as a branch of artificial intelligence is both an opportunity and a challenge.

    In the intelligent manufacturing process, machine vision mainly uses computers to simulate human visual functions, that is, to extract, process and understand the image information of objective things, and finally use it for actual detection, measurement and control.

    Yishi Zhitong CEO Huang Bufu believes that machine vision defect detection is a major "training ground" for artificial intelligence. Yishi Zhitong's high-precision visual dispensing system integrates the visual perception, motion control and dispensing execution functions of the dispensing process, and can be easily integrated with various actuators to form a terminal dispensing machine product in one step to meet various production requirements. The demand for line dispensing can also evolve from single-machine intelligence to multi-machine interconnection through deep learning.

    In addition, equipment failure monitoring and early warning is also a major application of artificial intelligence in industrial scenarios. This type of program can supervise every robot in a factory building and predict abnormal conditions of the robot, adding it before the robot has a problem. Send technicians to perform maintenance operations.

    In addition, if a robot fails, this type of program can also allow neighboring robots to automatically assume the tasks of their production line, so as to avoid or reduce equipment shutdown losses.

    Change is inseparable, and most of the application scenarios of artificial intelligence in the manufacturing industry are similar or related to the above. Industry insiders agree that the combination of artificial intelligence technology and robotics technology will change the pattern of the traditional robotics industry, just like the subversion of traditional mobile phones by smart phones.

    With the wings of artificial intelligence, can domestic robots bend overtaking?

    When it comes to industrial robots, everyone will inevitably mention ABB, KUKA, FANUC, and Yaskawa. According to industry analysts, the conditions for oligopoly are:

    First, the expansion of the market space is not fast enough to accommodate more similar manufacturers to enter, and the production capacity of a few large companies has basically met the total demand of all customers;

    Second, the technology is very mature and it is difficult to produce disruptive new technologies. It is difficult for companies in a catching position to achieve "curving overtaking" through technological breakthroughs.

    For domestic robots, international giants have always been in a state of catching up. Under such a market structure, domestic robots have begun to choose to enter from subdivision fields, trying to win in the local battlefield through "one skill". If China wants to change the situation of catching up, there are two major opportunities for transcendence:

    First, China is a huge incremental market for robot applications.

    Statistics show that in the 3C field, China’s annual output of mobile phones exceeds 2 billion, and the output of TVs, refrigerators, and air conditioners ranks first in the world; in the fields of logistics and e-commerce, the number of express parcels exceeds 40 billion, that is, 30 pieces per capita, ranking first in the world; in the field of food and chemical industry, the output of chemical fertilizers is ranking first in the world. A huge amount of actual industrial demand provides a huge training ground for the implementation of artificial intelligence.

    Second, China's talents and technology are in the first echelon.

    Compared with robot ontology technology, China is relatively leading in the field of artificial intelligence, which is reflected in the number and quality of papers published in the field of AI are among the top two in the world; it has made important contributions to the infrastructure of deep learning ; Well-known research institutes and universities belong to the first echelon in the world; they have swept the rankings in various AI competitions.

    In specific practice, with the improvement of domestic robots' cost performance, the industry's recognition of domestic robots has increased, and robot companies have creatively applied artificial intelligence technology and robotics to the actual needs of specific industries or application scenarios. There is huge space for proposing solutions and realizing corresponding products, which is also the key direction of entrepreneurship and innovation.

    However, the "curve overtaking" road will inevitably not be flat. Shao Tianlan pointed out that to truly enter the new era of AI + robots, Chinese robots still face challenges, such as relatively shallow accumulation in trajectory planning, compliant control, etc.; they need to compete with the Internet, autonomous driving, face recognition and other fields for super-class talents. In addition, long-term investment in all aspects requires great determination and ability.

    Similarly, Blue Fat Robot CEO Deng Xiaobai gave the industry a "vaccination shot": concepts and stories are easy to tell, but things are not easy to do. What can be achieved is an ideal, and what cannot be achieved is a dream. He believes that in terms of hardware, craftsmanship takes time to accumulate; in terms of software, the research and development and education of robotics software lag far behind Europe and the United States. "China has a market and there is hope, but there is a long way to go. Whether it is robots or artificial intelligence, it needs to steadily land the application of market segments and then expand horizontally." Deng Xiaobai said.

    Is the "artificial intelligence + manufacturing" on the tuyere a real boom or a carnival before the bubble burst? The answer to this question is probably that artificial intelligence that can be successfully implemented will produce great value; and the narrowly defined "sky castles" that rely solely on AI algorithms or technologies will not be able to adapt to the industry situation and will soon see the collapse of the bubble.