As early as March, Huawei announced the HUAWEI ADS 2.0 high-end intelligent driving system (HUAWEI ADS 2.0), and the AITO M5 Huawei high-end intelligent driving version (HUAWEI M5 intelligent driving version), the biggest improvement of this product is the new lidar and its corresponding high-end intelligent driving system, that is, HUAWEI ADS 2.0.
On the eve of the auto show, the intelligent driving version of the M5 equipped with Huawei HUAWEI ADS 2.0 was officially launched, priced at 279,800 yuan to 309,800 yuan. What will HUAWEI ADS 2.0 bring to the M5 Smart Driving Edition? What are the changes brought by HUAWEI ADS 2.0 compared to HUAWEI ADS 1.0?
Li Wenguang, President of Huawei's Intelligent Driving Solution Product Line and General Manager of Intelligent Driving (left)
In the face of these problems, at the Shanghai Auto Show, we interviewed Li Wenguang, President of Huawei's Intelligent Driving Solution Product Line and General Manager of Intelligent Driving, to further understand everyone.
At 9 a.m. on April 18, the first day of the Shanghai Auto Show Media Day, the official opening time of the exhibition hall, as the earliest batch of interviews at the beginning of this auto show, AutoLab and Li Wenguang's interview was held at the booth of AITO.
In the casual chat before the start of the official interview, Li Wenguang broke an "insider" to AutoLab, talking about the front lidar just released by Hesai before the auto show, Li Wenguang said that Huawei had this idea many years ago, but in the end did not choose this solution, or Huawei believes that lidar on the roof is a better solution, more conducive to its performance, although the aesthetics are affected, but just like Apple Smart Island, everyone will slowly adapt.
Insisting on doing what is right, rather than blindly compromising with the trend of terminal consumption, perhaps this is the self-confidence of a leading manufacturer like Huawei that is different from other people. In the interview, about lidar, AutoLab also raised a question that everyone is concerned about: "Now there is a discussion about whether to use lidar." Is it 4D millimeter wave and the like, may also have a flat effect, what do you think of this topic? ”
Li Wenguang: "These are two technical routes, a technical route of pure vision, and the other is multi-sensor fusion. The technical route of pure vision itself has its upper limit, and it is much lower than the upper limit of fusion perception. On the purely visual route, Tesla should be the best in the industry. But if the upper limit is stuck there, no amount of data can go up. ”
"Tesla now uses vision for inspection, but it is very difficult, the introduction of lidar will bring a lot of benefits, our lidar is easy to perceive the three-dimensional information of the object." Depth information is easy to have, detection distance is also accurate, from the effect of our training, the ceiling of multi-sensor fusion is obviously higher, we use a small amount of data, to achieve a short time of rapid improvement, the experience is obviously better than him, like Tesla FSD, many processing scenarios, obviously not as good as our processing effect. ”
Indeed, at the perception level, the capabilities of the M5 intelligent driving version are full. It is equipped with 1 overhead lidar, 3 millimeter-wave radars, 11 high-definition cameras and 12 ultrasonic radars, of which lidar can accurately measure range, the perception environment is realized by forward-looking binocular cameras up to 800W pixels, and its millimeter-wave radar also adopts new glass antenna technology, equipped with 4 MIMO antennas, the signal is stronger.
Stronger perception also makes it more confident at the level of algorithm decision-making.
In addition, for intelligent driving, perception "visible" and "understood" are not the same. In this regard, the highlight of Huawei's ADS 2.0 is also the use of BEV-aware network and the industry's first GOD network.
The BEV perception network allows vehicles to gain a God's perspective, and through training labels, use whitelists to identify obstacles. For example, through Huawei's rich scenario library of AI training clusters, HUAWEI ADS 2.0 can conduct simulation tests for tens of millions of kilometers every day, mine difficult scenarios, and use massive data to implement intelligent driving algorithms and continuous iteration and optimization of scenario strategies.
However, the problem is that all kinds of special-shaped obstacles in the traffic environment are theoretically endless, no matter how large the amount of data training, it is theoretically impossible to achieve label recognition of every one. For human drivers, even if we don't recognize certain obstacles, we know whether the area ahead is navigable.
The significance of the GOD network is that it no longer needs to identify and mark all obstacles in the traffic environment, but only needs to plan which areas of vehicles can drive, so in theory, even if the obstacles in front are outside the BEV perception network whitelist, the GOD network can still "identify" and avoid.
Tesla FSD's Occupancy Network occupies the same concept, but the difference is that Tesla FSD is a pure visual fusion solution, while Huawei's GOD network is through the perception of multi-sensor fusion such as lidar, millimeter wave radar, and camera, and in the case of bad weather and insufficient light, the perception and recognition ability of obstacles will indeed be superior.
And to go to HD map, plus lidar, HUAWEI ADS 2.0, its cost is lower than ADS 1.0 set?
In response to this problem, Li Wenguang made it clear that the cost is decreasing.
Also, what about the difference between HUAWEI ADS 1.0 and 2.0 experience? AutoLab also raises this question.
Li Wenguang believes that HUAWEI ADS 2.0 is based on perception as the core, so in the processing of many scenarios, there is no need to do a lot of change processing, and the mechanism of the map (high-precision map) is often not as good as the perception system.
Indeed, without the "guidance" of high-precision maps, with pure perception as the route, the vehicle no longer needs to "follow the map", and it will indeed give people a more natural intelligent driving experience. In fact, the more important significance of going to HD maps is that it can be quickly rolled out in different cities. In the interview, referring to high-definition maps, Li Wenguang also said: "One of the biggest advantages of no map is that generalizing cities is very fast. ”
The goal of HUAWEI ADS 2.0 is to rely on HD maps in the early stage to achieve the implementation of urban NCA in 5 cities in Q2 this year, and to achieve the landing of 15 cities without maps in Q3 this year, and the landing of 30 new cities without maps in Q4, reaching the NCA landing of 45 urban areas. If you continue to use high-definition maps, limited by the high-definition map updates that may occur every few months and the collection of longer periods, the speed of smart driving landing is much slower than the current goal of 45 cities in HUAWEI ADS 2.0.
The interview also talked about the HUAWEI ADS 2.0 parking experience.
At present, HUAWEI ADS 2.0 supports 160+ parking scenarios, and also supports cross-floor valet parking in the parking lot, when the owner completes the first autonomous parking, the next time he arrives at the parking lot, activate the valet parking assistance function to roam to the target parking space automatically parked.
In the interview, Li Wenguang, president of Huawei's intelligent driving solution product line and general manager of intelligent driving, also said: "PROBABLY NEXT YEAR, THE HUAWEI ADS SYSTEM WILL SUPPORT THE PARKING SUMMONING FUNCTION. ”
In fact, on HUAWEI ADS 1.0, AutoLab also experienced its automatic parking function, and the Extreme Fox Alpha S HI version equipped with this system performed very well.
AutoLab Super Out-of-Box Automated Parking Test Ranking
At the end of the interview, when talking about the business model of intelligent driving, Li Wenguang gave two dimensions: safety and time.
Li Wenguang believes that the demand for the entire intelligent driving is to solve the user's safety problem, and the other is that if the time on the road can really be liberated, the user will pay. This is just needed from a security and time savings perspective. Demand is the most important, as long as an industry has demand, everyone will make rapid progress in promotion. As for how much is this worth? Some people say that this is equivalent to replacing the driver, how much is that driver, at least I think that the price combined with intelligent driving ability now, it still needs to be taken step by step.
At present, the standard functions of the HUAWEI ADS 2.0 advanced intelligent driving system software package equipped with the M5 Intelligent Driving Edition have covered 19 functions such as high-speed LCC, urban LCC, and high-speed NCA. The optional high-end package also includes urban NCA, AVP (Automated Valet Parking) and urban LCC enhancements, with a one-time purchase price of 36,000 yuan, an annual subscription price of 7,200 yuan, and a monthly subscription price of 720 yuan.
Finally, let's take a quick look at the M5 intelligent driving version equipped with HUAWEI ADS 2.0, the highlight of which is that it will be the first to realize high-speed, urban high-end intelligent driving functions that do not rely on high-definition maps.
In hardware, it has sensors such as overhead lidar, visual perception HD camera, millimeter-wave radar and ultrasonic radar; In terms of software, relying on Huawei's ICT technology advantages, BEV network plus GOD network are adopted at the converged sensing level, and based on the multi-sensor fusion sensing system, it has a road topology inference network and anthropomorphic intelligent driving algorithm, which can realize high-end intelligent driving solutions that do not rely on high-definition maps.
It can be said that as a representative of the current technology to empower the automotive industry, HUAWEI ADS 2.0 will not only empower the M5 intelligent driving version, making it one of the top intelligent experience in the current market, its low cost and large-scale rapid landing ability will also promote the development of domestic intelligent driving a big step.