Meituan hopes to coordinate delivery vehicles, drones and riders to create a man-machine collaborative distribution network integrating air and ground.
Author | Tetsu Ta
Edit | Wen Liang
On December 10, 2021, the 4th Global Intelligent Driving Summit hosted by Lefon & New Intelligent Driving was officially held in Shenzhen.
This time, Leifeng Xinzhi Driving, with the theme of "Intelligent Driving Battle Moment", handed the microphone to 19 benchmarking companies in the industry, radiating 13 major technologies/scenarios, covering multiple dimensions such as intelligent driving algorithms, chips, perception, and landing, and only selecting the most representative enterprise in each field.
Following the two gold standards of "basic theoretical technology innovation" and "industry solution landing", the speakers shared with the industry their summary and review of past experience, predictions of future trends and sharing of effective models.
At the summit, Xia Huaxia, general manager of Meituan's unmanned vehicle distribution department, delivered a wonderful speech entitled "Intelligent Driving Practice in Urban Delivery Scenarios".
Xia Huaxia pointed out at the meeting that because of its small size and low speed, automatic delivery vehicles often give the illusion that the technology of the outside world is relatively simple, but when the urban open roads are driving at a speed of more than 20 kilometers, the uncertainty of pedestrians, bicycles, and motor vehicles on non-motorized roads often brings unexpected safety problems to automatic delivery vehicles.
Although autonomous driving technology is generally standardized, autonomous delivery vehicles mostly drive on unstructured roads with narrow roads, multiple obstructions, and many uncertainties.
He believes that if urban distribution is to achieve large-scale landing, it is necessary to realize the dehumanization of no safety personnel in the front line of operation. Now, on the one hand, industry players need to continue to improve autonomous driving technology, so that the entire automatic driving perception, planning, control, and positioning of the surrounding area are more safe and reliable. On the other hand, it is necessary to follow up the operational capabilities of manpower, maintenance, emergency safety personnel, etc.
In April this year, Meituan released the 20 magic bag of automatic delivery vehicles produced in accordance with vehicle regulation standards, which has undergone a number of tests such as vehicle performance test, endurance test, and adaptability to severe cold and heat environments, with a maximum speed of 45 kilometers per hour.
At present, Meituan's automatic delivery vehicles have accumulated more than 500,000 kilometers of public road delivery in Beijing's Shunyi District, and delivery orders have exceeded 100,000.
The following is the full text of Xia Huaxia's speech, leifeng network new intelligent driving has done not change the original meaning of the collation and editing:
Hello everyone! Just now Teacher Zhu completed a very wonderful sharing, which mentioned that unmanned driving is very difficult, and I deeply feel the same way. Because meituan has been doing unmanned driving for more than five years now, from the beginning to the present, we have made some progress every year, and we are also facing new challenges, and my own inner awe of this matter is getting bigger and bigger every year.
Today, I take this opportunity to synchronize with you on the one hand, the progress of Meituan in automatic driving in 2021, on the other hand, I also want to discuss with you what are the challenges of doing automatic driving in complex urban scenarios.
This picture is the application context of autonomous driving, divided into different scenarios and speeds. According to the speed and complexity of the scene, we draw many scene applications on the map, represented by different colors. Blue is the vehicle carrying people, red is the bus of urban public transportation, orange is the unmanned delivery field, and gray is the application of special scenes.
As you can see, the higher the right corner, the higher the speed and the more complex the scene. From the perspective of autonomous driving technology, the stronger the capabilities of the autonomous driving technology that we think it needs.
If you look at unmanned delivery in urban areas, although the speed requirements are not particularly high, the complexity of the scene is particularly high.
These diagonal dotted lines represent some of the contour lines of capabilities, and unmanned distribution may represent the need for our most complete autonomous driving technology in the most complex scenarios, which is the goal we have been striving to break through.
Logistics is generally divided into trunk logistics and urban distribution. Several guests mentioned self-driving trucks this morning, which are trunk logistics. Urban distribution is generally delivered at a speed of 40 or 50 kilometers or more than 20 kilometers. Among them, the logistics scenario is generally called branch distribution. There are also some in closed scenes such as parks, campuses, residential quarters, etc., and there are also many distribution needs.
Meituan now mainly focuses on urban scene distribution, which will involve urban roads, parks, residential roads and so on. The participants in this type of scene are very complex, and we hope to eventually achieve the combination of urban scenarios, including parks, delivery vehicles on open roads, and drones in the air, and collaborate with our riders through unmanned equipment to create an air-ground integrated human-machine collaborative distribution network to support the distribution needs at the end of our city.
In April this year, we officially released a new automatic delivery vehicle , the Magic Bag 20 , which is 1.1 meters wide, 2.5 meters long and has a top speed of 45 kilometers per hour. In terms of vehicle intelligence, we have added more sensors, greater computing power, and the total number of sensors in the entire vehicle has reached more than 30. In terms of vehicle hardware, we produce, manufacture and test according to vehicle standards. Among them, we have completed the vehicle performance test, endurance test, cold and heat environment adaptability and other tests.
Our cumulative total mileage of public road delivery in Beijing's Shunyi District this year is about 500,000 kilometers, and the cumulative delivery of real user orders is about 100,000 orders, and this year we have participated in the anti-epidemic work in Shenzhen, Guangzhou, Nanjing, Chengdu, Xiamen and other cities.
Automatic delivery vehicles should recognize traffic lights on public roads, and in non-motorized lanes, they should interact with many vehicles, pedestrians, old scooters, bicycles, battery cars, retrograde pedestrians and vehicles. When the overall operating speed is more than 20 kilometers and there are many battery cars retrograde, it will bring great challenges to the passage of unmanned delivery vehicles.
Automatic delivery truck is a delivery car, the driving speed is relatively low, in many people's imagination, is it relatively simple? But when we actually do it, we find that in this scene, we will encounter a particularly large number of "fireworks" in many cities, as well as unexpected long tails and complex situations.
I have listed a few pictures here, such as on motorized roads or non-motorized roads, pedestrians, bicycles and many other elements intersect, and many vehicles are parked on the side of the road, including when we are driving on non-motorized roads, many trees block the satellite signal very badly.
The entire autonomous driving technology is very consistent, including positioning, perception, decision-making, and control, but each link will face challenges. For the more specific challenges, I'll show you some examples separately.
Let's start with perception. In the current scene, because we have to drive on the highway, on the non-motorized road. The three graphs represent different cameras, and combined with the camera and radar, we can identify many obstacles around us.
From this picture, you can see that there are many different obstacles in the road, its speed is different, the direction of movement is different, and the types are very different.
For example, there are a large number of pedestrians and bicycles on the narrow roads of the park, how do we identify them in such a scene? If there is a bicycle parked here, we need to judge whether there is someone on the car or someone on the side of the car, whether it is a stationary car or a riding car, and so on.
Another is positioning. Many times we feel that positioning is a relatively solved problem, but in some complex scenarios in cities, there are still many challenges to positioning.
On the one hand, many scenes will degrade positioning, the so-called "degradation" is its surrounding scenes, the first possible satellite signal is blocked.
On the other hand, if you look at the surrounding environment visually, the environment is always unchanged. For example, in a long tunnel, no matter where it goes, the lidar vision matches and sees the surrounding tunnels are the same. Or on the revolving ramp of the basement, it's hard to know exactly where it is now without carefully identifying the information. For example, trees can block a lot of satellite signals, so that we have to use more integrated technical means to locate. For example, urban roads will frequently encounter a large number of urban road construction scenarios, the environment will often change, once it changes, we need to update the high-precision map in a very timely manner to make the vehicle better.
Let's look at decision control. In an urban environment, there are many elements of the road that interact with automated delivery vehicles. If the automatic delivery vehicle is driving on the highway, the trajectory and speed of the surrounding cars are relatively easy to predict, because most of the cars are going straight ahead, and occasionally changing lanes will gradually come over. But when there are many pedestrians and bicycles around, when many people are retrograde, or when many pedestrians cross the road on the road, how to predict every element around us, or where it will appear in the next three or five seconds, will greatly affect our decision-making and control actions.
The figure on the far right is a U-turn that is not very easy to complete by the automatic delivery vehicle on unstructured roads such as park roads, and if a large-scale U-turn is made, the requirements for vehicle technology are very high, because the planning and speed control of the path must be very well matched to achieve. We have now realized the three-dimensional planning of space-time integration, combining path planning and speed control in a model, in order to better make the automatic delivery vehicle travel stably in the drastically changing path. This is the challenge we face in decision-making control.
Then there is the security aspect. Vehicles must be safe on the road, and they must be safer than people. This security is divided into active and passive security. Active safety means that the vehicle should be able to avoid obstacles on autopilot; passive safety is how to minimize the damage caused to the surrounding environment in case of a real collision, whether it is my reason or someone else's reason. For example, when a person hits, the car reduces the damage caused by the collision through the shape design and material selection.
At the same time, we also consider data security. Our vehicles will collect a lot of data, a lot of it is more sensitive geographic data, we need to ensure that the transmission and storage of this data is relatively safe.
Another example is the network, whether it is the uploaded data or the control signal of the vehicle, it is transmitted through the network, how to ensure data encryption, and the immutability of the control signal, etc., these must be considered.
There is also functional safety. For vehicles, in addition to the mechanical structure of the vehicle, there are a large number of different sensors, and there is a very complex distributed computing platform. Each of these sensors, computing platforms, and software can fail. When some parts fail, the vehicle can not achieve fault tolerance, if it can not drive normally, it is also necessary to minimize the harm caused to traffic.
The five security areas of active security, passive security, data security, network security, and functional security are all discovered and solved one by one. So we now think from three dimensions: vehicle design, the entire system aspect including the software system, and the entire operation process need to be better designed so that our distribution can be safer.
Finally, looking to the future, we feel that if urban distribution is finally landed on a large scale, it must be dehumanized, and dehumanization is that there is no need for security personnel to follow in the front line of operation. If a self-driving car wants a person to follow, the economic cost is not even if it is not passed, and the same is true for unmanned delivery.
There are no seats on the automatic delivery car, but when the automatic delivery car currently lands in most cities, according to the policy requirements, a person needs to be followed behind the car, as well as a remote monitor. We hope that in the end, regardless of technology or policy, automatic delivery vehicles can drive on the road without front-line operation safety officers, focusing on remote monitoring to ensure operations. Therefore, on the one hand, we need to continue to improve the automatic driving technology, so that the perception, planning, control and positioning of automatic driving around the environment are more safe and reliable.
On the other hand, we are working the operational side, and if there are some emergencies, we can quickly keep up with the operation of manpower, maintenance, emergency security personnel.
In terms of policies and regulations, we need to work with the government to improve the ability of automatic driving, and the government can have more confidence and open up more right of way and more flexible conditions to the industry.
Finally, in terms of cybersecurity, the entire industry also needs to constantly explore.
I hope that through the efforts of Meituan and my peers here, as well as many friends, we will work together to make the autonomous driving technology, the entire industry, and laws and regulations continue to mature. Our goal is to reach every corner of the world through unmanned delivery, and we believe that this goal will definitely be achieved, but we hope that more people will help to make this thing happen sooner rather than later.
I hope you all continue to work hard!
Previous articles on the 4th Global Intelligent Driving Summit
The University of Macau must be loyal: "Conservative" can not promote the progress of the intelligent driving industry | The 4th Global Intelligent Driving Summit
2021-12-15
Su Shuping, General Manager of Innoviz China: LiDAR is a tool for intelligent driving vehicles, not to "show off wealth"
2021-12-16