On April 18, Jidu officially released the design details of the front part of the first car robot concept car, and the design of the headlights and the lidar at the front cover of the car were clearly visible. So far, Jidu's first front cover and forward dual lidar autonomous driving technology scheme has been exposed for the first time.
Seeing this program can't help but make me sigh that now the lidar is not only more numerous, the location must be unique enough, otherwise there is really no novelty, this shape makes me first think of the AE86 folding headlights, it is reported that the mass production version of the model lidar will also use this way that can be exposed and hidden, as to whether it is lifting or folding has not yet been announced.
But the flattering point is that compared with those placed on the "top of the head" or embedded in the front of the car, this shape has a retro style, as if it is a combination of the future and the past, and the most cutting-edge technology is displayed in front of us in a retro form, which is quite visually impactful.
What will we do next around the establishment of Jidu for more than a year? The media communication meeting revealed what the latest progress, and for the technical aspects to do a popular science analysis, I hope to let you have a more comprehensive and in-depth understanding of Baidu and Geely jointly built Jidu automobile.
01 More than 400 days after its establishment, what has Jidu done in a hurry?
Jidu is an automotive robot startup initiated by Baidu and strategically invested by Geely. The company was officially established in March 2021 with Xia Yiping as CEO. Seeing this origin, we know that Jidu can be said to be born with a golden spoon, with the Endorsement of the Internet giant Baidu and the domestic car leader Geely, Jidu attracted many eyes at the beginning, coupled with the new concept "car robot" thrown out at the beginning, it seems to create a subversive intelligent electric vehicle in form and experience.
Where is the disruption? To sum up, there are three points: 1. Free movement 2.Natural communication 3.Self-growth. Free movement is well understood, that is, this car robot can achieve L4 level automatic driving function, covering the whole scene of high-level automatic driving, and truly release humans to hand over the car to intelligent hardware devices to control; secondly, natural communication means voice semantic interaction, barrier-free communication, let your car become your confidant partner, make the car more humane, the previous complex physical buttons will no longer exist, and what functions you want to communicate with the car at any time can be realized The third point of self-growth is to continuously iterate and upgrade according to the habits of the drivers and passengers, so that your car understands you better, just like a personal butler to help you cope with various road conditions.
We know how the development cycle of traditional cars takes three years, but how does Jidu shorten from design, development to final mass production to less than two years? In fact, a very important point is the decoupling of software and hardware, the set of the vehicle's software algorithm, the vehicle's electronic and electrical architecture, and the vehicle body and chassis and other components to achieve dual-track parallelism, and even the development of intelligent software systems in front, greatly improving the efficiency of research and development.
Earlier we introduced the background of Jidu is Baidu and Geely, then in the shell of traditional vehicles, Jidu's auto robot will be built using Geely's SEA haohan platform; at the intelligent brain level, Baidu has a wealth of automatic driving data accumulation and intelligent cockpit research and development experience, so the development rhythm of Jidu will exceed other car companies several times. It can be said that it stands on the shoulders of giants.
Then according to the name of the set, the carrier carrying the brain and the shell is the set degree SIMUCar (Software Integration Mule Car), which has been iteratively upgraded to version 2.0.
Let's take a look at what has been done in the more than 400 days since its inception.
On March 2, 2021, Jidu Automobile Co., Ltd. was officially established, and Jidu officially opened the research and development of automotive robots.
On June 1, 2021, Jidu held its first press conference since its establishment, telling about the positioning and planning of Jidu Automobile.
On June 10, 2021, the photo of the first Jidu automobile robot sludge model was exposed.
On July 7, 2021, Baidu founder Robin Li visited Jidu and participated in the confirmation of Jidu's first product design.
In August 2021, Jidu and Continental and Hesai Technology cooperated to jointly promote the landing of intelligent car robots. The first automotive robot will be equipped with Hesai Technology's lidar
On September 24, 2021, THE SIMACar was launched to develop the front of the autonomous driving and smart cockpit.
On November 29, 2021, it signed a contract with Baidu and Qualcomm to launch the first domestic mass production model using the 4th generation Snapdragon auto digital cockpit platform - 8295.
On December 31, 2021, Jidu successfully integrated high-speed domain and urban intelligent driving.
At the beginning of 2022, Jidu launched a brand vision including the brand logo "Pixel-J", JIDU AI-Purple, and the exclusive "language" (ROBO Font) for automotive robots.
And at the 2022 CES show, it was announced that Nvidia's autopilot chip Orin will be used, which will also be installed on the domain controller of the "three fresh" ACU developed by Apollo.
Until now, the first large-scale brand launch date originally planned for April will be postponed due to force majeure. Previously, Jidu has announced that it will release its first automotive robot concept car in 2022 and unveiled it at this year's Beijing Auto Show. Its first mass-produced automotive robot will be launched and delivered in 2023, which will have high-level autonomous driving capabilities and bring users a high-level autonomous driving experience in multiple scenarios. At present, Jidu has started the research and development and pre-research of follow-up models, and its second mass production car will be expected to be released at the Guangzhou Auto Show at the end of this year.
Through this schedule, we can see that Jidu has indeed competed for every second this year, starting late but Jidu has innate advantages and acquired efforts, so in this wave of Internet car manufacturing, Baidu seems to be the first to hit the line.
02 Set degree intelligent driving technology solution first exposure, is dual lidar enough?
Lidar can be described as a cliché, we have also more than once popularized the working principle of lidar and its significance to the current automatic driving, each car company is also in full layout, trying to gain an advantage in quantity, and even play less than a few there is no need to say the slogan, this lidar "inner volume" trend seems to be gradually prevailing. But the strategy of this set is relatively rational, two lidar on the car, both to achieve safety and redundancy, but also to take into account the cost control, let's talk about what dry goods were said at the communication meeting, and whether the two lidars were enough.
First of all, Dr. Wang Weibao, head of Jidu Intelligent Driving, shared the latest progress of Jidu automotive robots. Jidu SIMUCar (Software Integrated Analog Prototype) has been iteratively upgraded to version 2.0, and its automatic driving system has been added to the lidar sensor, and integrated with the high-level automatic driving intelligent architecture JET1.0 (electronic and electrical architecture + SOA) developed by Jidu. In terms of hardware, at present, the world's highest computing power on-board chips Nvidia Orin and Hesai AT128 semi-solid-state lidar have been put on the car and participated in the test of the jidu automatic driving system. This means that the high-end autonomous driving technology solution for mass production, which is based on Apollo's unmanned autonomous driving capability, has been fully exposed for the first time.
It is worth mentioning that the pure vision and lidar autonomous driving scheme currently being tested and developed in the SIMUCar 2.0 stage is an independent dual system that can achieve "true redundancy" after mass production. Compared with the traditional solution, the two sets of automatic driving solutions of Jidu are mutual backup and complement each other, and the two sets of solutions can not only drive themselves efficiently, but also efficiently coordinate.
So back to the key lidar, we mentioned earlier that the location of this arrangement is very unique and innovative, so how can the specific parameters of lidar perform, can it help Jidu automotive robots to really achieve safety redundancy? The two lidars equipped with the Jidu automotive robot are AT128 from Hesai Technology, which not only has a super telemetry ability of 200 meters (@10% reflectivity), but also has a dot frequency of more than 1.53 million points per second, reaching a global high resolution of 1200x128, and every detail can be clearly depicted on the point cloud, providing a strong perception of the assisted driving system.
Compared with the 120-degree FOV designed with a single radar, the front double radar can achieve horizontal 180-degree FOV coverage, and in the driving scenes such as "ghost probes" and left and right obstructions, the recognition ability of left and right pedestrians or obstacles is stronger. Secondly, the identification of the key area of the vehicle is more accurate, in the area of 60 degrees FOV directly in front of the vehicle, the dual lidar can be doubled and overlapped, and the point cloud data on the target is more, and the recognition accuracy is higher. Third, in terms of the redundancy safety of high-level automatic driving, dual lidar can be mutually safe and redundant, which is more reliable than the single lidar solution.
Therefore, with the blessing of these sensors, Wang Weibao believes that the SIMUCar 1.0 stage is like a person's primary school stage, which is a key period for laying a good foundation, focusing more on the basic capacity building of automatic driving, including the development of software architecture and related basic functional logic, and adapting to the chassis. SIMUCar 2.0 is like the middle school stage, which is a period of gradual improvement and formation of basic capabilities, in this stage, the underlying architecture of SIMUCar has been upgraded to a high-level autonomous driving intelligent architecture JET (JIDU Evolving Technology) developed by jidu, including the electronic and electrical architecture EEA and the vehicle operating system SOA. At this time, the relevant domain controllers and sensors have gradually reached the mass production state, and some basic automatic driving capabilities, including ADAS functions, parking, intelligent interaction, etc., will gradually reach the mass production state.
03Lidar and visual perception, a mountain can accommodate two tigers?
After talking about the dry goods at this gathering exchange meeting, I can't help but want to discuss whether lidar and visual perception are stronger or weaker. Are the two complementary or mutually exclusive?
Let's first talk about the basic principles of autonomous driving and find out what role lidar and cameras play. Autonomous driving gradually transfers the driving ability and driving responsibility of the car from people to the car, which mainly includes three core links of perception, decision-making and execution.
Among them, the perception link is equivalent to the human eye and ear, mainly through the vehicle camera, lidar, millimeter wave and other types of vehicle sensors in the process of driving to complete the perception of the environment and vehicles, collect the surrounding environment data and transmit it to the decision-making level;
The decision-making process is equivalent to the human brain, mainly through the operating system, chip and computing platform to connect the received data for real-time processing and output corresponding operations and command tasks;
The executive end is equivalent to the human limbs, and the received operation instructions are executed to the vehicle terminal parts such as power supply, direction control, and lamp control.
Therefore, it can be seen that lidar and cameras are the functions of human eyes to help vehicles recognize the world.
So far, Weilai, Xiaopeng, Ideal, BAIC Jihu, Great Wall, SAIC Zhiji, SAIC R, Lucid, Volvo and GAC have all made it clear that they will mass-produce models equipped with lidar.
The advantage of lidar is that it has high accuracy and long detection distance, and can achieve higher accuracy than the camera in some extreme weather and at night, preventing misjudgment of vehicles and improving the redundancy of safety.
At present, lidar is divided into three categories: mechanical, hybrid solid-state and pure solid-state. These three categories have their own advantages and disadvantages, such as the most mature mechanical development, the scanning speed can be 360 degrees of all-round detection, but because of its large size is difficult to load mass production, the cost is also a big problem; hybrid solid state cost is low, suitable for large-scale mass production but limited field of vision, 360 degrees will not work, pure solid state will be the future development trend, there are OPA optical phased array and Flash two technical routes, but also wait for further breakthroughs in technology to achieve mass production. At present, the MEMS route in the hybrid solid state will be the mainstream of lidar development and the easiest to achieve front-loading mass production.
The data obtained by the visual solution is more similar to the real world perceived by the human eye, with the characteristics of light hardware and heavy software.
Due to the low price of the camera, the visual solution has a significant cost advantage and is easier to pass the vehicle gauge test. In addition, the image data obtained by the camera is more similar to the real world perceived by the human eye, and the morphological closest to human driving, and the high-resolution, high-frame rate imaging technology also makes the perceived environmental information more abundant.
However, the camera's perception in dark environments is limited, and its accuracy and safety have decreased. And because the visual solution in the context of reduced hardware requirements, its requirements for software are significantly improved, that is, it needs to rely on powerful algorithms to ensure the efficiency of image processing, command issuance and processing.
It can be seen that lidar and visual cameras have not been completely separated at this stage, and multi-sensor redundancy is currently a more conservative development route for various manufacturers. Therefore, in a car, the realization of multi-sensor fusion two-legged walking should be the most secure at this stage.
04 written at the end
Through this online sharing meeting of Jidu, we not only have more expectations for Jidu's automotive robots, but also let us see the focused and efficient completion of the set goals, and good products will also be presented. In addition, the integration and elimination between various technical routes will make the probability of safety closer to 100%, so we look forward to the visual algorithm can further break through, become closer to the level of the human eye, the lidar industry is more vigorous, so that the price continues to explore, more in line with the requirements of front-loading mass production, look forward to the concentration of locomotive robot mass production version can really appear when we can surprise us, let us wait and see.