Wen 丨guang Cone Intelligence, author | Liu Jian
Recently, car manufacturers have all brushed together and began to "do not do business". On September 7, Xiaopeng Automobile ecological enterprise Pengxing Intelligent released the first intelligent machine horse.
Earlier, Xiaomi, which was first a "10 billion car", launched the robot dog "Iron Egg" on August 10. Then, on August 19, Tesla exhibited the humanoid robot Tesla Bot at the AI Open House.
Although many netizens ridicule Xiaopeng to some extent is a bit of a "refers to the dog as a horse", but the main business of the machine horse and Xiaopeng to build a car, it really can not be said to be irrelevant.
Musk, who has always been fast," said tesla as a robot company on wheels. On August 18, Baidu founder Robin Li also defined his products as "automotive robots". The head men of these cars have treated cars and robots as the same thing.
On the 9th and 10th, at the 2021 World Robot Conference, John Hennessy, a Turing Award winner, former president of Stanford University and current chairman of the board of directors of Google Alphabet, pointed out that intelligent learning and iteration of robots in image recognition, sensor fusion, handling uncertain situations, etc., have become a key path for the evolution of self-driving car technology.
Whether it is a car, a person, a dog, or a horse, although the form looks diverse, in fact, many of the underlying logic of intelligent technology are connected. That is to say, whether it is a car or a robot, if you want to "recognize the road", you must rely on automatic driving technology.
<h2>The common fusion of cars and robots</h2>
The origin of cars and robots may be that Autobots leave Cybertron in search of new energy sources. In 1984, the large American animated series "Transformers" was released, opening the door to the fusion of cars and robots.
At Universal Beijing International Cinema, which was in trial operation at the same time as the World Robot Conference, the light cone intelligent editor also felt a "fire source battle" of a car man in advance through 4D movies.
(Universal Beijing Transformers Base, image from Weibo)
At last month's Baidu World Congress, Baidu gave a realistic answer. In addition to the Apollo unmanned vehicle can not deform the dry frame, automatic driving, human-computer interaction and other elements are readily available.
In addition, Ali unmanned logistics of the small wild donkey, meituan food delivery robot, although named robot, but in addition to the different manned carriers, in the field of automatic driving, many of the software and hardware technology of cars and robots are common, are based on visual perception and other multi-system integration of the scene cognition and decision-making.
Li Yuhao, CEO of Maiyan Intelligence, who focuses on community service robots, believes that "the highest form of robots is to do the same as people. "Perceive like a human, make decisions like a human.
He Xiaopeng, founder of Xiaopeng Automobile, said that the company is building intelligent robots with the model of building intelligent cars. Its own robotic horse has a car-grade intelligent driving system that can intelligently identify the environment in which it is located, and act and interact.
Lu Wenliang, Special Research Institute of the Automotive Industry of the Industrial Science and Technology Center of the Chinese Academy of Sciences, believes that "from sweeping robots to bionic robots to automatic driving, it is a gradually upgraded technical route." ”
Autonomous driving technology can be understood as three stages, car assist person (L2 ~ L3), human assist vehicle (L4), L5 level unmanned. The highest level of autonomous driving is to realize a robot to replace people to drive, which is the ultimate dream of all autopilots.
However, the L5 technology with a complete replacement for people, as well as the distance to the large-scale landing on the car, make practitioners feel that the future is uncertain. Even the confident Musk admits that self-driving isn't that easy. Even Su Zhen, who was the president and chief architect of Huawei's intelligent driving product line, is more pessimistic and believes that L5 automatic driving can never be achieved.
But the technology will not stop developing, the road to step by step, the meal to eat one bite at a time, L5 can start from "low speed". According to the grading standards set by SAE International (International Association of Automotive Engineers), the core technical problem of L5 autonomous driving is to achieve all-weather, all-regional unmanned takeover. Low-speed unmanned driving, such as unmanned logistics, is more suitable for different environments and easier to achieve in the identification of people and things. Accumulate technology for automotive driverless cars.
Musk has shown that Tesla Bot can share accessories including visual solutions with Tesla cars.
Zhang Chunhui, co-CEO of Zebra Zhixing, also said: "At present, the automatic driving of specific scenarios such as low-speed logistics is accelerating the application of landing, and after scale, it will have a huge driving effect on the components, sensors and other industries. ”
<h2>Speed reduction without descent, the difficulty of achieving L5</h2>
The low speed will help the all-weather, all-regional autonomous driving technology to land and complete the accumulation of L5-level technology. But at the same time, complex scenes such as communities and hotels make robots need to adapt to more realistic needs, which can be said to be "speed reduction without descending".
(Mai Yan intelligent self-cleaning sweeping robot "Aurora One")
Compared with the closed park of unmanned logistics and unmanned minibus, the community scene is more complex, closer to the road environment without traffic lights and mixed people and vehicles. In a sense, the accumulation of technology in actual complex scenarios helps to improve the overall capability of autonomous driving. Before the official realization of the car L5, the robot became one of the most potential scenarios for the L5 to land and even mass-produce.
Li Yuhao mentioned in the interview that Mai Yan Intelligence has conducted research on 23 needs in community scenarios and found that the highest demand and the greatest difficulty in the community are cleaning, followed by security, and the third is takeaway into the community. Behind these major problems is the mixed personnel in the community, the behavior of people is unpredictable, and the road conditions are very diverse. In the community robot cleaning work, it is necessary to adapt to the complex road conditions of the variety; it is also necessary to track the movement path of the avoidant and the pet at the same time; and to distinguish which are debris on the working surface and which are useful items such as ID cards and documents. This requires robots to have a strong ability to recognize the environment, people, and objects.
For unmanned vehicles and robot autonomous driving technology, in order to achieve a high degree of intelligence to work like a human, it is necessary to always carry out the "soul three questions" for yourself:
Where am I
Where am I going
How I got past
"Where am I": Whether it is a robot or a self-driving car, it needs to have the ability to locate it first. Through its own various sensors, such as lidar, millimeter wave radar, ultrasonic radar, car cameras, night vision systems, GPS, gyroscopes, etc., to determine their own position in a certain coordinate system, and correctly perceive the surrounding static and dynamic obstacles, and even establish a surrounding environment map, this is the robot environmental perception technology, it can be said that the environmental perception ability is the basis for the realization of all automatic driving functions.
"Where am I going": L5-level autonomous driving systems need to have the ability to make their own decisions and plan. For example, for delivery robots, the safe delivery of goods to the correct location is the decision-making goal, for cleaning robots, as full as possible coverage of the ground is the decision-making goal, for self-driving cars, the choice of the optimal travel route is the decision-making goal.
"How do I pass": When the automatic driving system knows its current location and the position to be going, it should consider how to pass, this is the third important technical content - path planning, obstacle avoidance and motion control, the goal of path planning technology is to choose an optimal path between the current location and the target location on the known map as a motion trajectory, obstacle avoidance technology is to use the robot or unmanned car on their own lidar, millimeter wave radar, ultrasonic radar, A series of distance-sensing sensors such as on-board cameras and night vision systems perceive obstacles and circumvent them, and motion control technology is the final actuator to ensure that robots or autonomous driving systems can quickly achieve the above goals.
In general, the autonomous driving technology of automobiles or robots is mainly composed of four aspects: environmental perception, behavior decision-making, path planning and motion control.
In the real scene, in addition to the internal "soul three questions", automatic driving also faces the external "soul answer" and interaction with people. How to read instructions and effectively feedback is a key factor in whether machines can replace people. Human-computer interaction, its interaction objects and interaction methods are not single.
In the driving scene, the driver doesn't just drive the car, but also interacts with people inside and outside the car. This interaction is often not in the form of language, and it is often a glance or an action to complete the gesture. For example, nod or wave to pedestrians to let them go first, or passengers get out of the car to pick up their luggage and knock on the body of the car to indicate that they can leave. The old driver replaced by automatic driving needs a tacit understanding of one eye.
At present, whether it is a logistics robot, a cleaning robot or a food delivery robot, it is also based on vision and speech recognition to achieve simple interaction, and the experience is mostly better than the intelligent interaction in the car, but there is still a long way to go from strong artificial intelligence.
<h2>Policies help the way, and financing does not make a mistake</h2>
On August 12, the Ministry of Industry and Information Technology issued the "Opinions on Strengthening the Management of Intelligent And Connected Vehicle Manufacturers and Product Access", which clearly pointed out that before enterprises implement online upgrade activities, they should ensure that automotive products meet the relevant requirements of national laws and regulations, technical standards and technical specifications and file with the Ministry of Industry and Information Technology, and changes involving safety, energy saving, environmental protection, anti-theft and other technical parameters should be declared to the Ministry of Industry and Information Technology in advance to ensure the consistency of automotive product production. Without approval, the automatic driving function of the car shall not be added or updated through software upgrades such as online.
Due to the issues of ethics, safety, laws and regulations, the implementation of the L5 in the car requires patient verification step by step. Liu Yinghang, a partner at Scale Partners, believes that L5 autonomous driving is an inevitable trend in the future, but it is difficult to achieve large-scale application in the short term.
Robots are different, China's current aging is serious, labor shortage, there is a large demand for robots. According to the census report, the proportion of China's population over 60 years old has reached 18.7%. According to the National Bureau of Statistics, China's birth population has declined for 5 consecutive years from 17.87 million in 2016 to 12 million in 2020. The reduction of the labor force and the aging need for labor care will form a scissor difference, and the labor gap will be further aggravated in the future. Compared with self-driving cars, robot landing has obvious advantages.
It is precisely because of the huge market demand that the market heat of robots is not inferior to that of automatic driving. In the field of robotics, the head of enterprises including Cloud Trace Technology, Gaoxian Robot, Damin Technology and other enterprises have reached hundreds of millions of yuan in financing, and the B+ round of financing of Damin Technology is more than 1 billion yuan. According to incomplete statistics, between January and May 2021, of the 81 financings completed in the domestic robot field, nearly 40 financing amounted to more than 100 million.
In the future, those who fly in the sky, those who run on the ground, and even those raised at home may be robots. Family communities, hotel restaurants in a variety of robots shuttle, L5 autonomous driving is around you and me. (This article was first published on the Titanium Media APP)