If you were to imagine, what would the ideal self-driving car of the future look like?
In terms of riding experience, of course, it is to maximize "freedom".
The driver's seat can be completely canceled, people in the car, the legs can be extended as long as they want, and they can lie flat with one button.
Lying down, the robotic arm will bring you a cup of tea.
The car is full of high-tech interactive screens, for meetings, games, movies, as you like.
Outside the car, the vehicle flexibly flashes and moves in the traffic, and with the blessing of cameras, millimeter-wave radar, and lidar, it perfectly realizes automatic driving and avoids various obstacles. All the cars seem to be in a huge neural network.
When putting the luggage, the robotic arm butler can do it for you, and we don't have to bother ourselves at all.
Coincidentally, the driverless concept car released by Didi yesterday is perfectly in line with our imagination.
The car is called Neuron, which means "neuron."
On the planet Pandora of "Avatar", trillions of trees are connected by roots, forming a huge network like a neural network, and then the "super intelligent" Eva was born.
In the absence of language, can it be perceived and shared with everyone? Eva has given us infinite inspiration.
Similarly, Didi's Neuron also embodies this concept everywhere - "empathy resonates, everywhere".
When L4 autonomous driving technology enters our lives, it will ideally provide a "hybrid network of humans and robots."
01 How the concept of Neuron was born
In the past 7 years, where did Didi's story begin?
Of course, it starts with a vision of the future.
In Paris in 1899, in preparation for the following year's World Expo, France organized a group of opinion leaders to imagine the world 100 years later.
In their imagination, in a world in the year 2,000, transportation could go up into the sky and the sea, carrying the whole family on its back.
At this time, the Wright brothers' plane was not yet in the sky, just 13 years after the invention of the first car.
While these visions are bold enough, they fail to capture two of the most important inventions in modern mobility: shared mobility and driverless driving.
Now, 120 years later, what kind of travel can we expect as passengers?
Didi's concept car, Neuron, paints an ideal future scene for us.
Eywa, a giant tree that can channel spirits in Avatar, can connect all clansmen, upload their consciousness, and feel, understand, communicate, and share without verbal communication.
This is Didi's archetypal cognition of services.
Hence the name of the concept car is DiDi Neuron.
Neuron is neuron, the smallest unit of perception in our body, allows us to empathize, and is ubiquitous in our body.
Empathy resonates everywhere
Where is the empathy and resonance of DiDi Neuron?
Although the overall length of the Neuron is "only" 4.5 meters, the wheelbase reaches 3.2 meters.
This is because this car no longer needs a driver, and after removing the driver's seat, the cabin becomes very large, and the passenger space is increased by 50%, which instantly makes the sense of prestige and luxury burst.
The eyes at the front of the car are sensor kits, the mouth is a triple screen on the front, and there is a joint screen on each side.
Why is the screen of the car so big? Because Neuron will also protect traffic participants outside the car - although there is no driver, Neuron's window will tell people outside the car what the self-driving is thinking.
For example, when someone crosses a crosswalk, Neuron will tell him: I will wait for you to pass, don't panic.
In short, whether the Neuron is turning left, arriving at the station, or preparing to start, people outside the car will see it clearly.
In addition to the 21 cameras, 8 lidar and 6 millimeter-wave radars in the whole body, the rest of the Neuron is also full of details.
For example, at the location of passengers getting on and off the bus, there is also a set of sensors so that Neuron will never stop in a puddle and muddy place.
The tail of the Neuron, unlike ordinary self-driving cars, is the noisiest, the most dissipated, the most wired, and the most ugly.
It leaves plenty of space and even has an "Easter egg" - a robotic arm.
It can help us with our hand luggage and all kinds of things, like a physical butler in the cabin. On the side, Neuron's design also reflects intimate and empathetic.
For example, the electric sliding door borrowed from the D1 design will not hit others when the door is opened.
The design of the top wing door is very ceremonial, so that passengers can get on the bus without bowing their heads. Even Yao Ming didn't have to bow his head when he came.
And in the cockpit, it completely gave us the experience of first class of the aircraft. After sitting in, the housekeeper will hand you water, and there are many flexible screens and interactive broadcast screens in the car.
02 There are no big models, only the real progress of Robotaxi
However, although the dream is released, Didi still has to return to reality, because there are still many questions that need to be answered.
At the open day, Meng Xing, the COO of Didi Autonomous Driving, quipped, "Today we do not pre-train multimodal large models, only the real progress of Didi Robotaxi productization."
In June 2020, Robotaxi first landed in Shanghai, and there were some problems driving on the road.
By 2023, in Guangzhou driverless car tests, Robotaxi will be able to handle a variety of scenarios with ease.
Safety is the first stage in the development of autonomous driving. In the past 1,020 days, Didi's self-driving safety index MPCI has increased 100 times, and there have been no major safety accidents.
Then the second stage is to consider autonomous driving services, from focusing on driving to focusing on passengers and operating experience.
Add the 24h family bucket
To this end, after consulting many passengers, it was found that the highest voice was "can the area be larger".
This demand reflects on the other hand, and many people want to integrate autonomous driving into normal life. In fact, taking Shanghai as an example, in the past three years, the coverage area of autonomous driving in Shanghai map has expanded by 7 times.
While expanding the map, the most difficult thing is how to solve the scene that Robotaxi has not seen.
For example, in a location in Guangzhou, a large number of parents' cars and passing students parked near the school after 4 o'clock one day.
For some common scenarios, Robotaxi can handle it flexibly. However, there will still be many scenes in real life that have not been encountered, such as traffic police on the road, unknown objects, and so on.
Therefore, the main challenges today are threefold: more traffic flow, more interactions, and more long-tail scenarios. In response to the above three challenges, Didi also gave its own solutions.
In the face of more traffic flow, which poses a great challenge to the delay of the Robotaxi system, Didi launched a perception large model and a prediction large model.
More traffic flow brings two problems, one is rational interaction, Didi introduced game theory decision model, which can capture the accuracy of all rational interactions.
The other is irrational interaction, which needs to introduce imitation learning to help automatic driving reference learn human driving behavior.
In the face of more long-tail scenarios, one type is the one that appears frequently, and you can actively mine the data you have already seen.
When encountering rare scenes, such as sudden plastic bags, predict and judge the trajectory of behavior by occupying the spatial network.
All of the above capabilities are the "urban generalization engine" that Didi has uniquely created in the past two years.
The second requirement is "can there be more sites".
For many people, they have encountered a dilemma, and the place they want to go is exactly between two bus stops, and they don't know whether to get off at station A or stop B.
Didi is to solve everyone's bus-like entanglement and compromise. In the case of autonomous driving sites, problems such as being occupied by other objects on the side of the road are often encountered.
In the past year, on a limited scale, Didi has increased the parking capacity of the Robotaxi by 96%.
But the desire to have more sites places a higher demand on Robotaxi's technical capabilities.
Hence the unlimited parking technology. It turns entire sections of the road into stops, and Robotaxi can park cars in any safe place, depending on the road conditions. This will be the focus of Didi's autonomous driving team this year.
Next, the third demand from many passengers is 24-hour service, how can Robotaxi be up to the task of working normally at night in the first place?
Traffic flow at night has dropped by 60% compared to daytime, but major accidents occur more frequently at night, mainly due to reduced visibility at night.
After Didi Robotaxi is equipped with infrared sensors, in the dark night, you can clearly see the crowd that the original camera could not see, so that the night service capability has been solved.
To achieve 24-hour continuous rotation, not only 24-hour phaseless operation is required, but even charging, maintenance, cleaning and other issues.
No, Didi's world's first autonomous driving automatic operation and maintenance center has been built in Shanghai!
The AGV trolley drags the Robotaxi to charge, and then the robot cleans the car, and finally completes a series of steps before delivering it to the door to continue operating.
KFC and Family Families have many stores that are open 24 hours a day, and now, Didi Robotaxi has also joined the "24-hour Family Bucket".
So, the question is, how far are autonomous driving services and ride-hailing services?
Didi did a test, and the results showed that it continued to approach ride-hailing in terms of send-off time, driving distance, user popularity, and call volume indicators.
At present, Didi's automation rate has reached 90%, and it will be able to reach 100% in the near future.
It can be seen that Didi Robotaxi can achieve such good performance, and it must be inseparable from the strong hardware blessing.
03 LiDAR + computing platform: key hardware for autonomous driving
The key components of autonomous driving are, of course, the eyes (lidar), the brain (computing platform), and the ontology of Robotaxi.
Adjustable resolution lidar
Think about it, when we drive, we are always in a state of watching and listening to all directions.
The three rearview mirrors on the left and right can be viewed at any time, the purpose is to grasp the road surface information at all times.
For autonomous driving, "eyes" and "ears" are equally important. Otherwise, if you don't even get the road surface information, how can this car drive.
Didi autonomous driving unmanned test in Guangzhou
It can be seen that in Didi's automatic driving, not to mention the cars driving on the road, even the crowd of people on the side of the road can be displayed in detail.
The lidar supporting L4 autonomous driving requires ultra-high point frequency, ultra-high resolution, ultra-long distance, and can adapt to various scenarios, which is quite demanding.
At the open day, Didi released Beiyao Beta LiDAR, which is the first 2K image-level high-precision LiDAR in China.
This lidar has a high point frequency of 3 million per second, ultra-512 line capability, a large field of view of 120° in the horizontal direction and 25.6° in the longitudinal direction, which can be said to be a "parameter explosion".
Simply put, it is to cover a large area, see far, and see clearly.
Traditional lidar harnesses of tens to hundreds of meters cannot see clearly at a distance of 200 meters, while ultra-512 lines can ensure the longest sensing distance of 200 meters. Even objects with low reflectivity, such as tires, perceive clear contours.
In the figure below, on the left is the relatively standard 100-line lidar on the market, and on the right is the situation of Beiyao lidar.
The test consisted of looking at 4 obstacles 200 meters away – picnic carts, people, bicycles, cone barrels, sorted from largest to smallest.
It can be seen that the general lidar has not detected the target yet, but there are already some point clouds on the 4 objects that can be seen. Because the point cloud is denser, it can gather more light and be brighter.
Point clouds are denser and therefore brighter
When the car goes further and reaches 140 meters, the point cloud on the right is already very dense, and the complete form of the person has appeared, and at this time, the bicycle and cone barrel in the left picture are almost completely without points.
When it reaches more than 80 meters, there is a point on the left, but it is not very clear, but at this time the display of Beiyao is already very clear.
At the same time, Beiyao Beta also has four different modes with adjustable resolution. It includes four modes: Normal Mode, Field of View Focus HD, Local HD and Global HD.
Field of view focus HD is Didi's original mode, the core meaning is that you don't need to see where you don't need to see clearly, sacrificing the resolution below in exchange for the resolution above, and the ability to adjust the ability to see farther places clearly.
There is also a similar function, called the local HD function, which can adjust our field of view (angle), which is equivalent to making a high-resolution window, and other places have made compromises, so that we can dynamically adjust the position of the HD window we want.
Another function is called global HD, which can instantly light up the global area, regardless of the region, and the resolution becomes high. This feature involves the ability to overclock and does not last long.
In short, the scenarios are different and the needs are different. Beiyao Beta can adaptively adjust the local or global resolution, and the perception effect can be best in different environments.
Orca killer whale computing platform
After talking about the most advanced lidar, let's talk about the newly released computing platform - Orca killer whale.
Why is it called killer whales? It is because it has a larger head and higher brain capacity.
As the first mass-produced three-domain converged computing platform in the autonomous driving industry, Orca Killer Whale is equipped with the on-board super central computing unit SCCU, which integrates the intelligent driving domain, intelligent cockpit domain and network connection domain, truly achieving low cost, small size and high integration.
The advantage of killer whales is that compared with the previous generation of hardware, its own cost has dropped by an astonishing 88%, the volume of vehicle space has been reduced by 74%, the number of core components has been reduced by 61%, and the number of wiring harnesses has been reduced by 33%.
The most intuitive advantage is that the space in the car has become larger, and the trunk can be put into an additional 20-inch suitcase. At the same time, the human-efficiency ratio of the whole vehicle has also increased by 7 times.
In a word, the cost has decreased, the size is smaller, and the performance has improved.
With the performance improved, the next step is production. At the open day, COO Meng Xing announced the latest progress in the mass production of autonomous driving.
At present, Didi autonomous driving is combining the capabilities and resources of new energy vehicle enterprises to define driverless new energy vehicles and put them into mass production.
Didi has cooperated with other companies in Robotaxi's product definition, model platform selection and cockpit intelligent driving system development.
It is expected that the first model of the project will be connected to Didi's shared travel network in 2025, enabling all-weather, large-scale hybrid dispatch.
It seems that the "sci-fi" scene of driverless cars speeding up passengers on the street is really not far from us.
In addition, at the technology open day, Didi Autonomous Driving announced for the first time that it has expanded its L4 autonomous driving passenger car solution to the field of trunk logistics and freight, and Didi Autonomous Driving Truck, an innovative business of Didi's autonomous driving, was officially unveiled.
04 The world's top L4 autonomous driving
Whether it is the commercialization of the landing or the speed of the overall progress, it is the confidence provided by Didi's autonomous driving technology.
So, why does Didi have to do autonomous driving?
First of all, it is the adherence to the same original intention - to make travel more beautiful.
Secondly, mobile Internet linked transportation has created today's travel industry, and Didi was born in the last decade and has grown rapidly.
In the next decade, the rapid iteration and commercialization of autonomous driving technology will be the next revolutionary inflection point in the travel industry, so Didi has to do autonomous driving seriously.
We must not only do autonomous driving, but also do the world's top large-scale, commercial L4 autonomous driving!
Zhang Bo, CTO of Didi Chuxing and CEO of Didi Autonomous Driving, said that it is still about 3-4 years to realize L4 autonomous driving technology, and Didi is also fully prepared.
It is worth mentioning that the full-stack core technology of L4 autonomous driving has been fully autonomous and controllable.
In the past 6-7 years of core technology research and development, Didi Autonomous Driving has mastered the full-stack core technologies of L4 autonomous driving, including perception, prediction, decision-making, control, large-scale simulation systems, and large-scale machine learning platforms.
In 16 years, Didi autonomous driving began to focus on research and development, and three years later, it was spun off into a separate department, and in August, self-driving cars debuted at the World Artificial Intelligence Conference.
In 20 years, the RoboTaxi project landed in Shanghai, opening autonomous driving services to the public for the first time.
In 21 years, the new generation of autonomous vehicle hardware platform - "Didi Gemini" was released, and in the road test in Shanghai, it did not take over for 5 consecutive hours.
In March this year, Didi autonomous driving officially "took orders" in Huadu, Guangzhou. As of April 13, Didi provides 24-hour autonomous driving services on 530 kilometers of roads in Shanghai.
The full-process operation center of Robotaxi includes four modules: order business center, security monitoring center, remote support center and automatic operation and maintenance center.
One of the main factors that Didi can achieve 24-hour Robotaxi service is that it has been doing online ride-hailing for many years and understands the importance of service experience and safety.
In addition, while accelerating the iteration of technology, Didi also creates an optimal software system on the basis of self-developed hardware. Not only is it reliable, but it also enables mass production, which in turn reduces costs.
So, what impact will autonomous driving have on the ride-hailing industry?
Previously, many voices believed that if self-driving ride-hailing is widely promoted, the manned ride-hailing market is bound to be squeezed. Even let the ride-hailing industry face a new round of reshuffle in the future.
However, in Didi's view, self-driving ride-hailing and manned ride-hailing are complementary, not substituting.
For a long time, self-driving ride-hailing will only be a complement to manned ride-hailing.
Moreover, new formats will inevitably create new jobs, such as the decade of ride-hailing and take-out banks, which have absorbed a large number of flexible employment.
Zhang Bo said at the open day that the latest version of the Didi Chuxing APP can be used in designated areas in Shanghai and Guangzhou to achieve mixed dispatch.
In the future, Didi autonomous driving will continue to make efforts to build a better autonomous driving service network for users and provide a better travel experience.