Lei Gang from The Temple of Oufei
Smart Car Reference Report | Official account AI4Auto
If the L4 autonomous driving scheme is cheap enough...
So what about progressive assisted driving?
Unpredictable unidentified scene? Need an emergency moment to take over the figure? Or is it a picture of the "owner's improper use" response after the accident?
The question is, how cheap can the L4 level automatic driving scheme be reduced?
10,000 US dollars, equivalent to about 70,000 yuan.
This is the latest answer sheet handed over by Shenzhen autonomous driving company Yuanrong, which Ali strategically led the investment of 300 million US dollars this year.
And this is only the beginning of the beginning, Yuan Rong revealed that the cost reduction action will continue, 10,000 US dollars has come, 5,000 US dollars, 3,000 US dollars will be far away?
$10,000 L4 Autonomous Driving Hardware and Software Solution?
Includes 5 solid-state lidar + 8 cameras, multi-sensor redundancy.
In terms of appearance, it has also begun to further iterate from "pill head" to "bangs".
Moreover, compared with mechanical lidar, solid-state lidar does not use mechanical moving parts, which is lower cost, longer service life, and easier to pass the car.
From the perspective of lidar suppliers, it is also converging with mass-produced smart cars.
However, to achieve "cost reduction" in the L4 automatic driving solution, the actual biggest challenge is not in the hardware, but in the system platform behind it.
Zhou Guang, CEO of Yuanrong Qixing, revealed that because of the long-term concern about cost issues, a series of innovative technologies have weakened the limitations of hardware on the system.
For example, Yuanrong's perception algorithm can adapt and fuse the data of solid-state lidar and camera very well.
For example, through the self-developed inference engine, complex autonomous driving systems can also run on low-cost, low-power computing platforms.
In addition, hardware devices such as self-developed cameras also reduce the overall cost of the autonomous driving system.
It is worth noting that this solution is also different from the L4 dimensionality reduction release usually said in the industry.
Because it is clear to the Yuanrong side, this aspect will be used in the front-loading mass production cooperation with the depot and its future Robotaxi operations.
In other words, one solution provides different applications, and one water source is available to the world.
The benefit of this "unified" approach lies in data processing and technology iteration.
Yuanrong Also made it clear this time that it will realize the continuous iteration of autonomous driving technology in the closed loop of data collection and analysis, algorithm improvement, simulation simulation, to real road testing and version upgrade through the data closed-loop system and data-driven way, and improve the processing capacity of automatic driving on long-tail scenarios from engineering.
This set of ideas for the implementation of the car and the iteration of data-driven technology is not without reference in the field of mass production vehicles and assisted driving.
Tesla and its FSD scheme are exactly that.
However, from the perspective of the real road ability shown by YuanRong Qixing, the Tesla FSD of the pure visual sensor may not be able to continuously take over if it currently challenges the Shenzhen CBD evening peak.
Challenging the Shenzhen CBD evening peak?
Along with Yuanrong's launch of this DeepRoute-Driver 2.0, there is also its latest actual road to show capabilities.
Because the headquarters is located in Futian District, Shenzhen, Yuanrong Qixing showed off the evening peak autonomous driving capabilities of a first-tier city CBD.
Yuan Rong said that in the 1 hour on the road, continuous automatic driving, the whole process did not take over.
There are main roads in and out of the ramp and in and out of the main way:
There are detours through temporary parking vehicles:
As well as coping with special Chinese road conditions such as second-round takeaways:
In short, in the whole video, the system of Yuanrong Qiqi shows the stability of the shenzhen CBD under the evening peak road conditions.
On another level, it also seems to be a silent proclamation:
If this system lands in mass production private cars, it will be a dimensionality reduction blow for all assisted driving systems.
L4 players fight back
All along, although all self-driving players have different route choices, the goals are actually the same, which can be described as the same destination.
We all hope to create mass-produced AI drivers to liberate humans from driving behavior.
The minimum goal is to assist the driver's route, providing intelligent assistance to the human driver in some scenarios, such as high-speed loops or parking.
The highest goal is the route of the self-driving player, and the whole scene and the whole time domain do not require human participation in driving, and really drive humans out of the driver's seat.
However, in the past few years, with the superiority of assisted driving in the upper volume, the system can be iterated through the owner's payment and data obtained after the road, so that this "progressive" route has become more and more recognized.
The most typical of these is Tesla.
And with Musk's personal appeal, even if Tesla is aggressive in the sensor solution and the iterative path of automatic driving, the development has not been fundamentally affected.
This further incentivizes more new cars, Tier 1, and self-driving startups to work with production cars and then take the data to iterate from assisted driving to autonomous driving.
And the corresponding Robotaxi targets players, such as Waymo, high and low, constantly being sung down...
Even if Robotaxi is considered to be the end of autonomous driving, the Waymo-style landing has been repeatedly questioned in terms of cost and data iteration.
More RoboTaxi players began to use the "de-allocation" method to achieve assisted driving capabilities in mass production vehicles, or reduce the number of sensors, or reduce the ability to reduce the allocation.
However, in the general direction, we actually hope to reduce costs and large-scale mass production.
Because the fundamental implementation path of the strongest AI driver still has to be the trinity of algorithm-data-computation, of which data is the most important in iteration.
In previous developments, the core problem of RoboTaxi players was the high cost of lidar...
But now, with the optimization of lidar in terms of mass production and cost, L4 level autonomous driving players have begun to appear technical "counter-kill".
Also figure progressive assisted driving what?
And this trend is not unique.
At the Shanghai Auto Show in April, Huawei's L4 level autonomous driving solution was installed on the Alpha S of Jihu Motors.
Subsequently, Baidu Apollo Moon was launched, claiming that the cost of the vehicle has dropped to 480,000 yuan.
(In addition to the vehicle itself, the cost of autonomous driving software and hardware is about 200,000)
On the side of the car company, it has opened an arms race around lidar, and the cost price has repeatedly explored...
As for today, Yuan Rong went further, and the whole set of software and hardware solutions was less than 10,000 US dollars.
And according to Yuan Rong, this is only the beginning, and the cost reduction action will continue, and may continue to advance in half.
So, the trend couldn't be more pronounced —
Thirty years east of the river, forty years west of the river.
After a few years of L2L4's ascendant faction, L4 self-driving players began to reverse their entry into the mass production car market as the cost of core sensors decreased.
The convergence of these two forces has become the biggest trend:
Smart cars.
It can be used as a private car or as a Robotaxi landing.
In the end, there is only one outstanding question left:
Who is more efficient in data processing and technology iterations?
Who is likely to be the biggest winner after the autonomous driving warring states dispute.
You tell me?
— Ends —