As we all know, the biggest application of industrial robots is in the production and manufacturing system, which plays a great role in the development of industry and also affects the healthy development of the national economy. Nowadays, China has accelerated the adjustment of industrial structure in the field of industrial robots, which has spawned many new opportunities and faces many new challenges.

'Three Great Mountains' in the new era

In the past, as China started late in the industrial field, it faced three big mountains—motors, reducers and controllers—but these technologies have been basically solved internationally. In recent years, China has also developed a number of strategies to study how to catch up with the international level. However, in addition to crossing these three mountains, we should look to the distance, think about new opportunities in the field of industrial robots, face new challenges, and how to break through this challenge to win the future.

Now, people think that industrial robots should be intelligent, then we need to know what the purpose of intelligence is and why it is intelligent. I believe that intelligence is nothing more than making industrial robots more convenient to use. So, how to make industrial robots more convenient to use, through research we have found the problem that restricts the realization of industrial robots.

First, programming robots is particularly time consuming. To be intelligent, the robot must first program it, that is, to write a machine language for the robot. The disadvantage of this method is that it is particularly time consuming. For example, in the traditional automobile industry, it takes three to five years for a car to be designed, and the robot programming cycle for producing a car is almost three to five years. However, now we have to apply the robot to an emerging field like mobile phone manufacturing. The mobile phone is replaced by a generation a year. As usual, it is necessary to write a new program for the robot once a year, so the cost of programming and use is greatly improved. . Therefore, we need to develop new programming methods to make the robot easier to apply.

Second, robot downstream companies cannot establish interactions between robots and the environment. While programming robots, companies must redesign new products to cope with the rapidly changing market. Sometimes, the production line should be reorganized, and the robots can be produced without installing them on the station. For the entire production line. It is a complicated process to say that robots have to interact with their surroundings. For example, to correct a robot, the robot has its own coordinates to be associated with the coordinates of the factory. This process is quite complicated. The downstream companies in the robot industry cannot implement this process. This requires the robot manufacturer to provide this service. Regardless of cost and usability, it is a big constraint on the development of robot application enterprises.

Third, the degree of integration between the robot and the sensor is not high. At present, robots are only simple industrial production, and can not be used in one machine. Our pursuit is to make robots more precise and intelligent, and to achieve multi-variety personalized production with high efficiency. To achieve this vision, the combination of robots and sensors is a key. Nowadays people talk about artificial intelligence. Artificial intelligence is first built on the basis of sensors, because touch is the first step in the development of artificial intelligence, which is enough to see the importance of applying sensors on robots.

Therefore, I have summarized these three major problems as solving the new challenges faced by the next generation of industrial robots. As long as these three problems are solved perfectly, the intelligent development of industrial robots will go further.

Make programming easier

First of all, talk about robot programming. Now everyone knows that traditional robot programming methods are particularly time-consuming to use, so can you develop more intuitive and convenient programming methods to efficiently program industrial robots in a short period of time? Therefore, we developed a programming method based on CAD model. How to understand? That is to say, if we use robots to manufacture products, we must first have CAD design models and drawings of the products. If the robot can directly understand the CAD model, then we can reduce a lot of programming process and turn the original manual programming into The robot's autonomous programming.

For everyone to give a 3D printing example, in the process of 3D printing, the robot holds a spray gun and sprays the glass fiber into a car compartment. The trajectory of the robot is very important during the injection because its speed and position determine how much fiber is injected into the carriage. Due to the different design of the car, some places are relatively thick, and some places are relatively thin, so the programming process is very complicated. If manual programming is used, it may take a week or two for this part to pass the test. However, if the CAD model is directly imported into the program, the computer can automatically generate the motion track of the robot. The entire program can be programmed in an hour.

Currently, this CAD model-based programming method has been applied in automotive companies. At the same time, it can also be applied to more complex production processes. For example, a 3D printing die is used. Printing automotive parts may only need to consider the distribution of the control material, and the process of printing the die is a process of spraying high-temperature molten metal, not only considering the distribution of the metal material, but also considering the temperature distribution, if there is a deviation, it will affect The mechanical properties of this product. Therefore, these factors must be taken into consideration when planning the robot trajectory. This is a very complicated process. Manual programming is almost impossible to implement, and CAD-based programming methods can achieve this process.

At the same time, this method can be based not only on the design but also on the sensor. We process a part to measure it. Based on this measurement, the trajectory of the robot can be generated and processed, such as sanding. We know that robotic grinding is applied in many ways, and programming is complicated, and the final program requires testing and polishing hundreds of parts, and using a robot with its own detection sensor is much simpler. The robot can detect the part where the part needs to be polished by the sensor, and then the computer automatically generates a polished track and re-detects and corrects the polished part after the polishing is completed. Therefore, we turned the original open-loop process into a closed-loop process, which not only reduced the programming time, but also improved the quality of the product.

Of course, there are many aspects to the application of robots, such as service robots. There is no CAD model in the programming of the service robot, which can interact directly with people. The communication between people is in the language, so it is especially important for the future service robot to let the robot understand the human language and directly program in the human language. However, since the emergence of artificial intelligence, how to make robots understand human natural language has always been a problem, then how to solve this problem? In fact, we have a big advantage, that is, the sensor inside the robot. These sensors can turn simple language understanding into a process of reciprocal feedback, extending the original low-level feedback control to the upper level and expanding into the natural language processing process. As a result, the language understanding of the robot becomes relatively Accurate, the programming process has become relatively easy.

From a traditional perspective, robot programming has become a more intuitive programming, and humans have told their robots through their language.

Robots should be 'self-reliant'

Then, talk about how to make the robot easier to set up with the environment and make it easier to self-correct. Many people simply think that after buying a robot, they can work instead of people. However, after buying the robot, it is very disappointing, because it is very troublesome for the robot to actually replace the person to work. The robot has its own control coordinates, and the user does not understand how to establish the connection between the coordinates of the robot and the coordinates of the factory. So after the robot is installed, how to make it aware of the environment, interact with the environment, and work in the same coordinates is very important for the user. A quick understanding of the environment is an important aspect of robots, which is also a sign of robot intelligence. Therefore, we have done some work in this regard.

In the past, there were two types of corrections for robots. One is to correct the relationship between the robot and the environment, that is, to combine the coordinates of the robot with the coordinates of the environment conceptually, and the other is to correct the coordinates of the robot itself. These factors need to be confirmed by correction, and are very complicated. It is difficult for users to make robot corrections, which requires robot manufacturers to do, which greatly increases the cost and time for robot users.

So why is it difficult for users to make robot corrections? In the past, the method of correcting the robot was based on points. We saved the trajectory of the space motion in the way of recording the points, and then calculated the coordinates of the transformation through these points. This method is quite complicated because we need to record a lot of points to calculate this coordinate. Now, we have developed a Line-basedCalibration method that directly records known lines. The line carries more data than points, which is much more convenient in the process of calibrating the robot. Therefore, we have installed a corresponding device on the robot, and the robot can perform the correction of the repeatability, which largely solves the user's problem.

Liberation of quality inspectors is not a dream

Finally, let's talk about how to combine robots and sensors to improve the intelligence of the robot.

In the production process, sometimes we have to measure the size of the two coordinates. The traditional method is to use a coordinate measuring machine to measure one point at a time, so that the cost is high. If we embed the sensor on the robot, the sensor can take a picture measurement on one area, so that the high-density 3D point cloud can be presented with low cost and high efficiency.

Taking the inspection of high-speed rail as an example, the railway company has to send a lot of vehicle inspectors at night to check the safety of the train, which is particularly time-consuming and laborious. Therefore, we are wondering whether the sensor can be directly detected by the robot and then compared with the train CAD model to find out the possible problems of the train. Therefore, we have specially developed a robot for detecting trains, which can flexibly extend to the bottom of the car, and take a large-area photo with a sensor to compare with the CAD model to quickly detect the entire car, not only efficiency. High and high precision.

Take the production of car seats as an example. In the past, the soft and hard comfort level of the car seat required the quality inspectors to perform touch detection one by one, and then scored a score through the subjective judgment of the quality inspector to judge whether the seat was qualified. As we all know, the auto parts are produced by the car service providers, so the mass production of seats must have a quantitative production standard, so the quality inspectors themselves have a large error. We use a robot-mounted sensor to detect more than a dozen indicators such as the softness and smoothness of the seat, reducing errors and improving efficiency. At present, our detection method has been used in the automobile factory in Mexico, and the data can be transmitted back to the United States, that is, remote detection is also possible. At the same time, such detection methods can also achieve the individual needs of customers. For example, customers can tailor a seat to suit their needs.

Through the specific case explanations of these three aspects, I want to convey to everyone the idea that in the field of industrial robots, we should jump out from the previous three technical challenges and see what new technical challenges will be in the future. How to solve these challenges makes the current robots truly become our ideal intelligent robots to help humans explore the unknown world.

(Source: Internet)