Autonomous driving, the final battle

As more affordable solutions emerge, the "second half" of the autonomous driving war has ushered in new development opportunities.

Since when, statements like "L4 level autonomous driving capabilities are reduced to L2 scenario usage" have become popular in the entire automotive industry.

Although we are not very willing to approve of the speculative marketing of "wearing a high-level hat but doing low-level things", such statements and practices have indeed opened up more room for operation for the development of autonomous driving, and have also further pushed the verbal unmanned driving to the "actual combat" stage.

There are apparently two reasons why high-level autonomous driving cannot be promoted: first, there are no specific laws at this stage to support the road rights of high-level autonomous driving; second, for car companies, there are "fatal" potential liability risks when high-level autonomous driving is on the road. 

Of course, whether true high-level autonomous driving can be achieved technically is still an unknown, so the field of autonomous driving at this time is more like a black box. Once opened, it may be a treasure chest or a Pandora's box.

All sectors, including commercial vehicles, passenger vehicles, and special-purpose vehicles, are trying to find the optimal solution that is more compatible with intelligent driving; new smart car concepts such as Robotaxi, Robobus, and Robotruck are also gradually becoming more common due to the hot market.

The results are obvious. After years of market testing, the large-scale implementation of intelligent driving for commercial vehicles is much easier and faster than that for passenger vehicles.

The Robotaxi-dominated track is home to a variety of players including Baidu, AutoX, and QINGZHOU Zhihang. The existence of these autonomous driving companies has opened up a path for the entire automotive industry, from the large-scale implementation of autonomous driving in commercial vehicles to the reduced-dimensional use of passenger cars.

However, whether this path is feasible is still unknown.

01. From results first to price war

People often use the large-scale implementation of autonomous driving as the key to distinguish whether autonomous driving is mature or not. However, according to the chasm theory, in order to develop from an early immature market to a mature market accepted by the general public, it is necessary to cross the large chasm that runs across the middle.

Now is no longer the era of "good wine needs no bush", the openness and transparency of information makes the strength of product competitiveness more apparent at the first time. Therefore, in addition to more advanced technology, the key point is also the control of cost.

Compared with other autonomous driving companies, AutoX can be regarded as a typical "results-oriented" autonomous driving company; in order to achieve high-level autonomous driving, it is even somewhat "inhumane".

Last July, AutoX released the fifth-generation fully autonomous driving system AutoX Gen5, with its luxurious configuration that was jaw-dropping.

In terms of perception hardware, AutoX Gen5 is equipped with a total of 50 high-definition automotive-grade sensors, including 28 8-megapixel automotive-grade cameras with a total of more than 220 million pixels per frame; the world's highest-resolution 4D millimeter-wave radar with an angular resolution of 0.9 degrees; 6 laser radars , including 2 high-specification 128-line laser radars , which can achieve more than 15 million point cloud imaging per second. 

In terms of computing platform, the latest AutoX XCU domain controller configured in AutoX Gen5 has a computing power of 2200TOPS, which is a new high in domestic vehicle-mounted supercomputing computing power. It is much higher than the 144TOPS computing power of Tesla's self-developed FSD computing platform.

If we estimate based on the hardware peripherals such as lidar and cameras required by Auto X Gen5, the cost of the entire vehicle has exceeded 2 million.

However, regarding this outrageous price, Xiao Jianxiong, founder of AutoX, said: "You can buy an assisted driving system for 200,000 yuan, but you definitely can't buy a self-driving car for 500,000 yuan. If a self-driving car dares to tell you it costs 500,000 yuan, I suggest you stay away from it."

Unexpectedly, as soon as the founder of AutoX finished speaking, Baidu Apollo cast a vote against it.

When it comes to the Robotaxi track, Baidu Apollo is definitely one of the earliest. Many industry insiders even call Baidu the Huangpu Military Academy in the field of autonomous driving. Yu Kai of Horizon Robotics, Wang Jin of Jingchi Technology, and the core founding members of Pony.ai and Leading Technology all came from Baidu.

After years of accumulation, Baidu's autonomous driving capabilities have blossomed in the Robotaxi track. However, unlike AutoX's "hardware stacking" approach, Apollo is taking the path of cost reduction.

Also last year, Baidu and Polar Fox jointly released Apollo Moon, which is dedicated to Robotaxi scenarios.

The car is built based on the Alpha T model of Polar Fox. Its sensor configuration includes 13 cameras, 5 millimeter-wave radars, and 2 laser radars, and also supports sensor self-cleaning.

In this regard, Wang Yunpeng, vice president of Baidu and general manager of the autonomous driving technology department, revealed the cost of the Apollo Moon vehicle and unmanned driving kit - 480,000 yuan.

To be fair, compared with the high price of 2 million, the cost of less than 500,000 yuan has already entered the range of most online car-hailing operating costs, and the commercialization of driverless cars is bound to take a big step forward.

But the pace of technological advancement will not stop. In less than a year, the price of autonomous driving solutions has been reduced by another zero.

On May 18, at its first brand day, QINGZHOU Zhihang launched an autonomous driving solution with a mass production cost of only 10,000 yuan.

When explaining the DBQ V4 autonomous driving solution, Yu Qian, founder and CEO of QINGZHOU Zhihang, said that DBQ V4 can support 1 to 5 laser radars, 0 to 4 blind spot radars, 6 millimeter-wave radars, and 12 perception cameras, achieving 360-degree perception capability without blind spots or dead angles, and the left and right sides are redundant.

In other words, this is an autonomous driving solution that can be "customized" for dimensionality reduction. The solution with a cost of 10,000 yuan mentioned above is the lowest configuration of DBQ V4.

As for why the cost can be reduced so much, Yu Qian, founder and CEO of Qingzhou Zhihang, explained: "Technological innovation and large-scale mass production have jointly brought about cost reduction. At the same time, with the support of Qingzhou Zhihang's underlying methodology, efficient technology iteration has also been achieved."

In fact, if we follow the logic of technological development, autonomous driving solutions should become more and more affordable. And looking at the deepest chain reaction, the charging system for autonomous driving software will also gradually collapse due to cost reduction.

02. How to have both soul and market?

As a market pioneer and leader, it is understandable that Tesla is used as a role model. Therefore, whether it is the development steps of autonomous driving or the marketing methods, many car companies will consciously or unconsciously move towards Tesla's direction.

First, regarding the capabilities of autonomous driving algorithms, the general consensus among car companies is that they should give priority to mastering them in their own hands; second, regarding autonomous driving chips, if the capabilities are sufficient, they should also go the route of self-research; finally, the IP of car companies is gradually overlapping with their founders.

It is not difficult to find that Ideal has begun to set up its own chip company, seemingly wanting to replicate Tesla's FSD; NIO has long since gotten rid of Mobileye's closed mode and embarked on the path of self-developed autonomous driving algorithms; Xiaopeng has been "self-developed algorithms + NVIDIA chips" from the very beginning, so in comparison, it has taken the lead and is ahead.

Of course, looking at the entire automotive industry, not all car companies have the ability to develop their own autonomous driving algorithms. Zhiji, BYD, and Lotus will all use Momenta's autonomous driving algorithms; WM Motor has a close relationship with Baidu and can rely on a big tree for shade; not to mention Zeekr, which is still trapped in Mobileye's vicious circle...

So an interesting scene emerged - car companies shouted "soul" on the one hand, and "the body is honest" on the other hand. In fact, to put it bluntly: for more market share, why not sell the soul?

With the development of technology, autonomous driving has gradually formed multiple systems, among which Tesla is the most unique, insisting on the pure visual technology route. As for other car companies and Robotaxi companies, most of them choose the hybrid route of "LiDAR + Camera + High-precision Map".

From the perspective of car companies, although they do not require their models to be extremely intelligent, they should at least not make it a shortcoming. From the perspective of consumers, although they do not necessarily have to take their hands off the steering wheel, they should at least get the intelligent driving they know.

Where there is demand, there is a market, so there are many opportunities for autonomous driving software and hardware companies. However, it is still important to note that opportunities and challenges often go hand in hand.

When it comes to having both soul and market, Tesla should be considered the first.

It can be seen that Tesla has chosen to take control of its own FSD chips and autonomous driving algorithms. Nowadays, 4680 battery technology has also begun to mature and mass-produce, further strengthening Tesla's dominance in vehicle manufacturing.