East Hangzhou Lin 'an City, west Litang Sky City, two places are 2400 kilometers apart. At a time when almost all the big Internet companies are doing the same thing -- using technology to get into the physical world -- something wonderful is happening in both places.
NetEase is the dominant player in this matter.
It loads its own game technology accumulated for many years onto the excavator body, so that the excavator can operate normally and efficiently when no one is in the cockpit. Artificial intelligence's sense of order and human labor division become more detailed, giving birth to new labor relations. Litang is the first real landing scene of NetEase Fuxi mining robot.
They regard the excavator with intelligent system as an intelligent agent. For an "excellent" intelligent agent, it must not simply repeat its past actions, but constantly learn and improve its own ability in task practice. Which means that the excavator learns on its own.
Man-machine collaboration is an eternal topic. In this case, we can clearly see the complexity and difficulty of NetEase's transformation of excavators. But more importantly, after years of evolution and development of video text, game text and audio text in China, NetEase now brings game technology to the face of the entity industry with its innovative spirit.
"Are you here to build the railroad?"
In September 2022, 66 years after artificial intelligence was established as a discipline, the world's first excavation robot to build a plateau railway operated here. From then on, an era of man-machine cooperation for extreme realistic environment landed on earth.
The excavation robot is installed at the construction site of Litang section of Southwest Plateau Railway. In some ways, there is a huge disconnect. Just come to the local people can not easily take a bath, otherwise easy to catch a cold caused pulmonary edema, life-threatening. The signal is weak, and 2G networks often block people's chat messages in swirling dialog boxes. The earth is high and the sky is cold, the heat is seriously insufficient, and the crop varieties are single. The industrial base is weak and there are no large industrial enterprises.
However, the imagination of plateau railway transportation has never stopped, no matter how difficult the geographical environment, in the face of human subjective initiative, can stop the heroic ambition of man's victory over heaven.
At the end of the 19th century, the British set out to build the Sichuan-Tibet Railway from northern India through Sikkim to Tibet, and then from Lhasa across Kangding to Chengdu. The attempt failed.
Litang, Sichuan Province, is a gateway fortress with an average elevation of 4,000 meters. It is known as the "World High City" and "Sky City". The last time Litang became known to the outside world, it was after the popular short video platform Douyin blew up a Tibetan man named Ding Zhen. Subsequently, attracted by the name of the tourists came, very lively.
However, from the perspective of railway construction, the railway belonging to Litang is called "the most difficult railway to build" because of its high mountains, cold and hypoxic conditions and extremely complicated geology. It was said in the West that it could not be completed in 100 years, but China aims to complete it by 2032.
Hardship must have its own unique value. Through the efforts of generation after generation, in September 2022, a team from the riverside city of Hangzhou all the way west, alternate means of transportation, finally arrived in Litang, Sichuan province. The NetEase Fuxi team, a top institution of NetEase specializing in the research and application of AI for gaming and pan-entertainment, came here to make final preparations for the field operation of the mining robot. As long as there is no one here to excavate, the world's most difficult reality will be treated as a generic problem.
One of the builders of the Litang project is China Construction No. 8. The full name of the Eighth Engineering Bureau of China Construction Co., LTD., has made great achievements in the construction industry at home and abroad.
A year ago, a team from the eighth Bureau of China Construction, which is in charge of building railway projects, finally settled on a specific construction plan for a section of the Plateau Railway after a long journey and precise measurements. Workers from all over the country rushed to build another miracle in the history of human transportation.
During construction, Litang County had thousands of excavators working side by side, but it was hard to find workers. Even though a worker could earn tens of thousands of yuan a month, there were few excavators working in Litang for a long time, and the turnover rate was high. Many of the workers who came to Litong to build the railway were not easy. They often suffered from severe altitude sickness such as headache, fatigue and vomiting within a short time. The high altitude also makes them worried about humpback, hair loss and irreversible organ growth, and many choose to leave.
On the first night of their arrival, the NetEase Fuxi team experienced a high level of nausea. Some people said that it was much worse than going abroad to suffer from jet lag. Some people also suffered from vomiting and headaches.
It is difficult to employ workers and difficult to construct. Construction efficiency is a long-term problem that puzzles the construction side in implementing local infrastructure construction tasks.
When the Fuxi team arrived in Litang County, taxi drivers, hotel owners and restaurant owners asked them the most: "Are you here to build the railway?"
Yes, not really.
They built the railroad in the same way as other workers, but in a very different way.
AOP is the essence of mining robotics
Dawa, an excavator master in his 20s, was born in Xiangcheng county, Garze, Sichuan province. When Dawa was young, his family was in poor condition and could only afford to send his sister to school. Later, he worked in a sand factory, gradually gained the trust of the boss, "let me learn the excavator".
Dawa, it seems, is a natural excavator adept, quickly on the road. In September 2022, he became the first master in charge of an unmanned excavator.
He does not need to be in the cab, but uses the smart handle to twist up and down, left and right, so that the excavator can dig in the direction of the handle.
"It's convenient, you don't have to stay in the cab, you're not tired." 'I was in the cockpit all the time, working 10 hours a day. It was tiring,' Mr. Dawa said. Young people tend to be more attracted to new things, "I like to try new things. My job is like playing a game."
At present, the excavation robot jointly developed by NetEase Fuxi and the eighth Bureau of China Construction Bureau is mainly used to pilot loading tasks at the mouth of the plateau railway tunnel. The intelligent operation mode of mining robot "one person and multiple machines" greatly improves the efficiency.
It is very important to understand the working environment of the mining robot. The eighth Bureau of China Construction Bureau took NetEase Fuxi team to visit the construction process of the tunnel operation -- first drilling the tunnel and blasting; Excavators inside to clear rubble, piled in a place; The loader constantly transported the gravel to the tunnel mouth to vacate the interior construction space; At the mouth of the tunnel, the excavator loads the gravel into the carriage, the truck transports the gravel to the downstream gravel plant for processing into sand and stone materials, the truck transports the sand and stone to the concrete processing plant, and finally is used for road construction. The whole construction process forms a closed loop.
Fuxi's team focused on more detailed work. During blasting, explosives are buried on the wall of the tunnel every certain distance. If the depth and amount of explosives are not arranged well, the blasting effect is poor, and more concrete is needed to reinforce the tunnel, which is very high cost. These experiences are from the exploration and summary of colleagues in the eighth Bureau of China Construction. Fuxi's team likened their experience to that of a mining robot. "The prediction of environmental plus blasting action can also be modeled and learned by AOP, just like the hydraulic model learned by excavators." The in-depth investigation on the front line also allowed the team to see more practical application scenarios of AI technology.
AOP is the essence of mining robot, including perception, cognition, decision, action four parts. Based on the historical data it collects and the simulation environment it learns, it can make predictive modeling more accurate than humans. Understand it, and you've got the heart of the mining robot.
Within NetEase, AOP is called agent programming. In short, it can have visual, auditory and tactile abilities similar to human beings. Through hardware equipment, it can imitate human hands, arms, mouths, legs and other parts of the joint, generate interactive behaviors with the environment, feedback information, and finally realize the process of self-learning.
It was difficult to understand what AOP was until I saw two excavators on a construction site in Baima Lake Ecological Creative City, Binjiang District, Hangzhou, and the vague concept was not matched to specific excavator parts.
Retrofit excavator
On a low, wintry riverside in Hangzhou, a computer and a desk stand next to a 22-ton excavator at the site of the NetEase Phase III construction project.
Liu Ming, a researcher, opened the computer and entered the robot remote control client interface. He clicked on the lower left corner of the behavioral command -- click where to dig, click where to press, click the ground, dig and dump, and the unmanned excavator in the cockpit would move freely.
To be honest, it's a bit cyberpunk surreal.
The mining robot project was initiated within NetEase in October 2021. Soon, they bought a small excavator model for more than 10,000 yuan to understand the real working principle. This is the key to control the excavator. In the working process of the hydraulic system of excavator, the hydraulic oil is sucked by the working pump from the bottom of the tank through the oil filter, and the hydraulic oil with a certain pressure is output from the oil pump into a set of parallel distribution valves.
In the initial transformation process, the key problem to be solved is how to let the algorithm learn the hydraulic model. The hydraulic model of different excavators is different, and the algorithm can achieve self-adaptation of different excavators.
"In fact, this is the process of imitating people. People get the excavator, as long as they move the position, the excavator will adapt to the new environment." Zhang said their underlying technologies are derived from Fuxi's intelligent cloud engine, AOP framework, and intelligent man-machine collaboration platform.
In short, the mining robot is a seedling from Fuxi's AI soil.
They went through three stages. In the first stage, instead of buying a real robot, you just need to write a game simulation algorithm in a virtual environment on the computer side. In the second stage, after buying the model, the oil circuit is transformed, the simulation algorithm is input into the excavator, and the excavator is controlled through the game controller. In the third stage, they bought a 7.5 ton excavator for practical operation.
They made detailed "human-like" modifications to the 7.5-ton excavator. The multi-way solenoid valve is like the cerebellum of the whole excavator, helping the excavator to maintain a balance of motion; A variety of sensors and controllers give the robot vision and touch, collect external information, lock the position and state of objects, and realize the remote intelligent control of vehicles.
The parts on the 7.5-ton excavator are very messy. Many of the parts are customized by Fuxi's team looking for suppliers. Most of the parts are only one in the market. They became much more skilled when installing parts on the 22-ton excavator.
There is a teacher named Zhang Qiang who teaches by example. He is in his 20s and left home to work after graduating from high school. He learned excavator technology on the construction site in Gansu Province, and then moved to Hangzhou, Zhejiang Province, and now directs the research and development of excavation robots in NetEase Phase III Engineering.
"Every excavator master has his own working style. Some people are very critical of excavator actions, and they will pursue the aesthetic feeling of some actions; Some people are more interested in efficiency and it doesn't matter how many buckets they have." Zhang Qiang said he is more pursuit of excavator work efficiency and overall beauty. To some extent, Fuxi mining robot also highlights these two points.
There is a consensus within NetEase Fuxi that man-machine cooperation is an upgrade of the understanding of AI. The intelligence of the machine needs to be carried by human beings and the intelligence of human beings needs to be assisted by machines. The development of intelligence should not make human beings quit. With the development of technology, there will be new jobs and new tasks, including Zhang Qiang's new work content.
Therefore, at the root node of the connection between the agent and the world, there must be no shortage of human beings, but the role may constantly change from the executor of the task to the organizer, manager, decision maker, or discoverer.
With the vision of exploring the world, they found in the investigation that the upstream and downstream production relations of railway projects are mainly: implementer, contractor boss, construction workers.
Among them, the contractor mentioned that if the excavator loaded gravel truck was not full, then he would have a lot of losses every day, he asked the excavator master to load fast and full. If excavators can load more trucks in a short time, they can make more money. Therefore, "efficient and full load" has become an important direction for the excavation robot team to transform the excavator.
In September 2022, they completed the transformation on a prototype of the same tonnage in Hangzhou, packed all the transformed equipment into several large boxes, and moved them to the Plateau Railway construction site intact. After nearly a week of installation, debugging and deployment, the excavation robot was finally put into use.
By means of automation control, man-machine cooperation and other technologies, the working efficiency of medium-sized mining robot in the dumping task test is comparable to that of a 10-year excavator master. But it is still early days, and there is still a long and impressive way to go.
Later, NetEase Fuxi team members have returned to the plateau. Flanked by high mountains, they were debugging equipment on an excavator in a ravine, holding spare parts in one hand and oxygen tanks in the other. They felt very uncomfortable, but they were very excited when they talked about unmanned mining robots.
"It was an uncomfortable and somewhat unique experience." NetEase Fuxi team members have debugging equipment twice in Litang, which is unforgettable.
"Game AI can solve problems in the real physical world"
Mining robot department is a very young business line of NetEase Fuxi, responsible for the development and application of mining robots. Its birth is closely related to the fate of NetEase.
In March 2016, AlphaGo, a GO program developed by Google's Deepmind team, beat professional nine-dan go player Lee Sedol in a stunning victory, indicating the great potential of AI technology in the game field. NetEase is determined internally to incorporate the latest AI technologies, such as reinforcement learning, into its flagship MMORPGs.
The following year, Fuxi Laboratory was formally established. After years of development, its research content has already included reinforcement learning, image action, virtual human, natural language, user portrait, big data and cloud computing platform, and has a number of industry leading technology solutions such as digital human, intelligent face pinching, AI creation, AI anti-plug-in, AI battle matching, AI competitive robot, etc. It has been used in more than 20 scenes of more than 10 internal games of NetEase, and has served more than 100 customers. The application has been called hundreds of millions of times a day. Researchers write and publish cutting-edge technologies into academic papers, gaining influence in academic circles and the game industry and feeding back to the lab.
But it is surprising that Fuxi can move from the virtual world of the Internet to the real physical world and solve important problems in the industrial sector. The transfer and exploration of technology make the mining robot more full of surreal digital style.
However, NetEase, a company with rich experience in real industries, ranging from pig farming to cultural tourism, has already departed from the narrative line of pure Internet companies and created a new, looming thread. In mid-to-late 2021, NetEase hopes to enter a more "real" physical world, but what space in the physical world? It is not clear.
The Fuxi team has also communicated with NetEase CEO Ding Lei many times about the idea that "game AI can solve some problems in the real physical world". "Every time we go in a new direction, every time we go into a new area, we are very careful." NetEase internal a familiar with the event said.
The birth of a commercial idea is often closely related to the general trend of China's economic development. Since the Internet economy has entered the era of stock competition, it has become a new trend of China's economic development from virtual to real. Internet commercial development with AI technology as the core will encounter the "last kilometer" problem. The reason is that this round of AI relies on "learning" in virtual environments, which are highly limited to data or simulation environments. With the Internet of Things, industrialization, 5G, virtual and real
