Real car experience + field visit, DJI vehicle test base unveiled for the first time.
Author | Wen Liang
Edit | Zhang Dong
Regarding the dynamics of DJI's vehicles, many people's impressions are still stuck at this year's Shanghai Auto Show.
At that time, the three major cross-border giants of DJI, Huawei, and Baidu - drones, ICT, and the Internet rarely gathered in the smart car track to participate in the exhibition as independent exhibitors and release related new products, full of highlights.
Among them, the "DJI Car" brand turned out to be a success, earning all the topicality and filling the expectations of the industry.
Although DJI rarely disclosed the progress of its in-vehicle business in public for more than half a year since, in the final month of 2021, the low-key player has a new move - releasing its first self-driving brand trailer.
Taking this opportunity, the new intelligent driving also made in-depth exchanges with DJI, trying to understand the underlying logic of its layout of intelligent driving and the vision they ultimately want to achieve.
In addition, we also went to DJI's test base in Longhua District, Shenzhen, for on-site visits and real-car experiences.
Test base visit
DJI's on-board test base is not far from the city.
After we entered the factory area, there was no "DJI Car" sign around. If it were not for the introduction of the staff, it would be difficult to know that the ordinary three-storey bungalow in front of it was its test base.
Since the test base prohibits taking pictures, the new intelligent driving is based on the memory of the day of the visit, trying to restore the main process of the test base to the greatest extent.
The key areas of this visit are: vehicle restructuring, simulation, perception calibration and other three major working areas.
Vehicle restructuring
The first stop of the new smart drive is to visit the vehicle restructuring area.
The process of vehicle restructuring is mainly divided into main line process and auxiliary line process. In the main line process, the first thing is to dock the received vehicle for basic inspection, including vehicle status, body size, etc.; at the same time, the auxiliary line process will check the parts that need to be used, including binocular/monocular/fisheye cameras, lidar, millimeter-wave radar, supercomputing platforms, etc.
The vehicles in the main line process enter the restructuring station after passing the inspection, and the corresponding components are dismantled according to the requirements (on the premise of not changing the appearance of the vehicle as much as possible), while the parts that have passed the inspection on the auxiliary line process are installed in the specified position, connected to the vehicle control system, and the intelligent driving system is initially formed.
However, the engineers on site also stressed that the specific rules and sensor configuration of the current restructuring can be changed with the needs of customers, and whether it is a fuel vehicle or an electric vehicle, the general restructuring cycle is 4 to 5 working days.
After completing the above steps, engineers need to debug the vehicle as well as the intelligent driving system to check for and resolve anomalies in the overall link to ensure system compliance.
Simulation
After the vehicle is restructured and debugged, the system will be connected to the simulation scenario and start the all-weather simulation test.
On the one hand, the system can form a structured description in the virtual world, thereby improving the stability of the system in real and complex scenarios; on the other hand, the modification system is integrated with the original data generated by the sensor model in the virtual world, and then compared with the real data, which can quickly optimize the modification algorithm.
DJI has also developed a simulator for engineers to manually manipulate vehicles in the virtual world.
Perceptual calibration
The last stop on the tour is the perceptual calibration area.
In addition to simulation, after the modified vehicle goes offline, it is also necessary to calibrate the sensor position and attitude information to ensure the final perception effect of the restructured system.
According to reports, the calibration station of DJI vehicle can be 1:1 forked according to the needs of car companies. The initial calibration of the intelligent driving system can be accurate to one or two millimeters, the angle deviation will not exceed a few degrees of zero, and the overall calibration process is about three minutes.
If the user encounters bumps and vibrations during the actual driving process, the system can fix the deviation problem through a self-calibrating algorithm.
After the successful passage of software tests and perceptual calibrations, the construction vehicle initially meets the conditions for participation in actual road tests.
On the day of the new smart drive's visit, there were only a few engineering vehicles in the parking area on the first floor of the test base, but the staff said that about 100 engineering vehicles were being tested outside at that time. In addition, there is also a maintenance area inside the test base to provide maintenance services for long-term testing of construction vehicles.
It is worth mentioning that in the operation area of the test base, as far as the eye can see, you can see a lot of grounded Slogans posted on the wall, one of which reads, "Cohesion to ensure delivery."
Perhaps in the promotion of DJI's in-vehicle business, "delivery" has become a very critical step.
Live on the road experience
The new smart driving test ride is a DJI vehicle, although it was not allowed by the video, we still remembered a few interesting small scenes and wanted to share them with you.
The real car experience of the new intelligent driving is nearly an hour or so, passing through the city expressway with clear lane lines and the main road of the city. Moreover, the system supports the destination navigation function, which can realize basic operations such as autonomous overtaking lane change and autonomous up and down ramps.
In order to "test" the sensitivity of the system, we deliberately issued a lane change instruction when there was a car in the next lane, but the system did not perform the change of lane, but the voice responded, "Lane change is currently prohibited." When you encounter a large truck and the other party has the possibility of changing lanes, the system will automatically choose to avoid.
At an intersection, we encounter a special-shaped car that has certain challenges even for human eye recognition. At that time, it could be seen on the interactive screen that the system did not frame the "strange alien object" in front of it, but still made a natural and smooth decision, stopping at the right distance and waiting for the green light.
Pictures taken by the staff
In addition, we passed a special scene with a separate ice cream bucket in the middle of the road, and the system also successfully identified and avoided.
However, for the sake of insurance, the current urban road under the U-turn, intersections, people and vehicles mixed traffic and other scenes still need to be manually taken over.
DJI Believes that the underlying logic of intelligent driving systems is the same whether it is a city or a highway. This is also an important reason why intelligent driving in different scenarios can be opened.
In the nearly one-hour experience, the overall feeling was relatively stable. Moreover, the sensor configuration of the new intelligent driving experience system is relatively "simple" - 1 binocular, 1 monocular, 4 fisheye, 5 millimeter waves; not equipped with lidar and high-precision maps.
DJI explained that what they want to build is an intelligent driving system that can be mass-produced, and mass production needs to consider cost. If you adopt the strategy of "stacking materials" during development, it will be very difficult to "go from luxury to frugality" in the future.
The old horse knows the way, which is another expectation of intelligent driving
On the second day after the new smart drive visited the test base, DJI launched its first self-driving brand promotional video.
It is also novel to say that intelligent driving must be inseparable from the car, but DJI car propaganda film does not really appear in a car, but with "horse" as the protagonist -
A horse with sufficient experience can make corresponding route planning and decisions according to its own perception and thinking to execute the instructions issued by the rider.
This is also DJI's imagination of intelligent driving.
Image taken from DJI's car promotional film "Letter Horse by Rein"
In terms of landing path, DJI did not choose a leap-forward route to cut into the market, but firmly did L2.
At present, it is not uncommon for models on the market to be equipped with L2-level functions, such as lane keeping and cruise control, but most of the production vehicles still follow the hardware integration method of distributed architecture. The problem with this approach is that there is an isolation between the sensor and the controller, resulting in inconsistencies between functions and functions, affecting the user's intelligent driving experience in different scenarios.
In other words, the industry has not yet achieved a driverless experience that can completely get rid of hands and eyes.
Therefore, DJI Believes that for mass production vehicles, only by making L2 and making it perfect can it be possible to achieve more advanced automatic driving.
In order to open up the intelligent driving function in multiple scenarios, DJI vehicles prefer to adopt a centralized architecture, based on a central domain controller and a set of sensors to make ODD continuous, so as to create a smart intelligent driving system that can be relied on like a horse.
However, it is not a simple task to do a good job in L2 and achieve the progress from L2 to L3.
As mentioned earlier, this year, the track of smart cars is very lively.
In addition to DJI, Baidu and Huawei have also repeatedly stressed that it is necessary to empower car companies and help car companies build good cars; from the perspective of landing paths, market positioning and product matrix, there is a certain intersection between the three. On the other hand, many players who started with visual perception technology, led by SenseTime, have also increased their layout efforts, and the heat of the track has added a handful of firewood.
So, what is the underlying logic of DJI's in-vehicle business? What are the advantages?
In fact, drones and smart cars seem to be incompatible, but the core problems that need to be solved are highly similar. Especially in terms of intelligent driving, the technical core of both is inseparable from perception, positioning, decision-making, control, etc.
In DJI's view, drones and intelligent driving belong to the same category of spatial intelligence, and it is not too difficult to migrate the accumulated technologies such as visual perception algorithms, multi-sensor fusion, positioning, obstacle avoidance, flight control and other UAVs to cars.
In addition, with the accumulation of space intelligence technology, DJI in-vehicle can greatly reduce the dependence on lidar, GNSS, high-precision maps and V2X, helping to reduce costs.
In addition to the above-mentioned software and hardware joint development capabilities, there are also manufacturing capabilities.
Different from the common FOUNDry model in the industry, DJI has a self-built parts factory, and DJI's supply chain resources and mass production experience accumulated over many years in the native field of unmanned aerial vehicles can also be used to a certain extent, which not only ensures quality, but also further reduces costs.
While paying attention to itself, DJI Has not stopped thinking about the development of the industry:
Although the landing path of each player is different, every car company wants to launch a product with a better intelligent driving experience. At present, DJI Has launched the science column of the official WeChat account, and has also opened up some traffic aerial survey data sets for industry people to learn and exchange. At the same time, DJI is also calling for more technical standards experts, policy-making experts and legal experts to join the field.
Perhaps, intelligent driving is not a standard technology, but a desire. DJI Vehicles knows that the ultimate realization of this wish will require more people to participate.
END
Who is the performance king of China's intelligent driving middle and upper reaches of the industry chain?
Autonomous driving, may not have entered the "second half"