Author: Zhu Yulong
As Robot Taxi, a once-popular self-driving car service, faced setbacks in its commercial launch in the United States, discussions about the future of the autonomous driving industry have resurfaced. The two major problems with this technology are the reliability of its operating systems and the high cost.
Recently, Baidu announced that the cost of its sixth-generation self-driving car, Apollo RT6, without a steering wheel, has been reduced to 250,000 yuan, which is only one-tenth of its competitors. With this business model, the price of autonomous vehicles in the future could be cut in half. For Baidu, self-driving cars have become the company’s best story in the capital market. As Baidu’s first growth curve came under pressure, people couldn’t help but wonder, how was the cost compressed and is this feasible?
How was the cost compressed to 250,000 yuan?
Apollo RT6 claims to cost only 250,000 yuan, which is nearly half the price of Apollo Moon, the fifth-generation self-driving vehicle launched in June 2021 (The complete cost of Apollo Moon, including the car and the self-driving kit, is 480,000 yuan).
First, let’s take a look at the car itself: Apollo RT6 adopts its self-developed “Apollo Galaxy” self-driving architecture, with 100% vehicle specifications and seven-fold redundancy. As for software, it is equipped with a fault diagnosis and risk reduction system that integrates the entire vehicle and self-driving capability. When any single component or system fails, the spare redundant system can be activated. In addition, the vehicle is equipped with 38 external sensors and a computing power of up to 1200Tops.
To achieve self-driving, it is necessary to rely on environment perception sensors to collect information about the road environment, send the collected data to the central processor of the car for processing, and identify obstacles, drivable roads, etc. Based on the identification results, the car speed is planned and automatic control of the vehicle’s travel is executed. The entire process must be completed in an instant and the delay must be controlled within milliseconds or even microseconds to ensure the safety of self-driving. To achieve instantaneous processing, feedback, decision-making, and planning execution, the requirements for the computing power of the central processor are extremely high. As the level of self-driving rises, the demand for computing power increases by at least ten times. Therefore, with the increased awareness of autonomous driving, the fifth level fully autonomous driving requires at least 500 TOPS of computing power per second.
# English Markdown Text
The cost in 2021 is still 480,000 yuan, but how did the new generation of autonomous vehicles manage to cut the cost down to 250,000 yuan? Baidu explains that it customizes its platform according to its operational scenarios and applies a proprietary intellectual property platform to significantly reduce the cost of its vehicles. How exactly was the cost compressed? We can roughly estimate:
From the perspective of large-scale procurement, we assume that through a joint venture company, a large amount of development costs can be shared, although it’s not impossible based on the current situation.
Based on the current situation, the battery cost, assuming 80 kWh and estimating at 1.3 yuan per Wh, requires 104,000 yuan, the electric drive system accounts for approximately 10,000 yuan per month, the overall cost of other high-voltage components is 6,000 yuan, and the entire set of electric vehicle-related systems is 120,000 yuan. The additional brake and steering redundancy is estimated at 5,000 yuan, plus the cool other systems cost 25,000 yuan, and the basic parts of the body (not related to autonomous driving) cost 150,000 yuan.
Now let’s take a look at the cost of the autonomous driving-related components:
The vehicle production sensor and computing unit used by RT6 have also significantly reduced the cost of autonomous driving kits. The equipped sensors outside the vehicle include 38 in total, including eight LiDARs. If we make an estimate:
Based on the 1200TOPS calculation board, there are four Orin chips, which cost around 20,000 yuan.
Eight LiDARs cost about 40,000 yuan, accounting for 20,000 yuan for long-range, and 5,000 yuan for short-range.
Plus other sensors, the entire kit costs about 15,000 yuan.
The overall estimated cost of the autonomous driving system is about 75,000 yuan, and with additional redundancy, it’s roughly within 100,000 yuan. In addition, China’s manufacturing advantages, supply chain, and mature production and manufacturing capabilities further reduce the cost of autonomous vehicles.
According to Baidu, the sixth-generation autonomous vehicle with a cost of 250,000 yuan can deploy tens of thousands of such vehicles across the country. In the second half of 2023, RT6 will be put into trial operation on Baidu’s shared autonomous vehicle platform, Luobu KuaiPao, and its annual production capacity will gradually increase from tens of thousands to hundreds of thousands based on operational deployment rhythm. The RT6 invested in Luobu KuaiPao by Baidu will have a steering wheel.
Besides cost, what else did Baidu not disclose?In June, Baidu’s Jidu Auto released its first mass-produced concept car, ROBO-01, which is also its first autonomous vehicle robot. A limited edition of their first mass-produced model will be released in the fall of 2022, which shares 90% similarity with the ROBO-01 concept car. Jidu Auto will unveil its second mass-produced model at this year’s Guangzhou Auto Show and begin accepting orders in 2023, with deliveries starting in 2024.
As we can see, Baidu’s subsidiary centers around the era of intelligent automotive 3.0, using cars as robots, with the starting point being the transfer of driving responsibility from humans to AI. AI drives automotive evolution, and the ultimate goal is autonomous driving.
Logically, Apollo RT6 currently reuses many components from Jidu Auto’s ROBO-01. This model showcases the exterior, interior, foldable steering wheel, autonomous driving, and intelligent cockpit. All physical buttons inside and outside the cabin are removed, and vehicle functions are controlled through voice command. Simultaneously, the 3D boundaryless integrated screen replaces the triple-screen, equipped with high-performance active lifting-type spoiler, U-shaped folding steering wheel, and more.
In fact, steering wheel folding is not yet allowed by laws and regulations, but it provides directional research for higher-level autonomous driving. Equipped with the Qualcomm fourth-generation intelligent cockpit chip 8295, and two NVIDIA autonomous driving chips Orin X, with computing power up to 512TOPS. It adopts a “vision + lidar” autonomous driving solution, equipped with 31 external sensors, including two lidars, five millimeter-wave radars, 12 ultrasonic radars, and 12 cameras.
According to my understanding, Baidu actually started from two directions this time: Apollo RT6 for Robot Taxi and the ROBO-01 concept car for consumers, which is a two-in-one design. Jidu Auto’s target market is family cars, with the first mass-produced car priced at around 200,000 yuan, directly targeting Tesla Model Y.
Therefore, from this perspective, a retail price of 200,000 yuan is definitely not possible, but 200,000 yuan may be the basic cost of the ROBO-01, hence the 250,000 yuan Apollo RT6 was born. As a relatively independent brand, Jidu Auto will create its own sales, delivery, and after-sales service network and will open its first brand-owned store this year. The vehicle will be mass-produced at Geely’s Hangzhou Bay factory.
Currently, Apollo has three business models:
-
Provide Apollo autonomous driving technology solutions to OEMs: which doesn’t look viable.- Car Manufacturing: By integrating Baidu’s autonomous driving innovation, Jidu Auto aims to bring the most advanced technology to the market as soon as possible.
-
Robotaxi for Autonomous Driving Taxi: This is the story we see now.
Summary: In fact, talking about it all the time is about putting things together. Under various unfavorable factors such as the epidemic, rising raw material prices, and supply chain pressures, achieving mass production and delivery, Jidu Auto still faces multiple challenges.
At present, Baidu has not strictly distinguished the requirements of Robot Taxi and household cars in several business models, and many contents are overlapping. In other words, the vehicles developed for consumers and the Robot Taxi for operations are combined, which is a heavy burden for the development team. The maintenance and use between the two will face different operating environments in the future.
Therefore, from the current perspective, whether it is Jidu’s target price of around 200,000 RMB or the cost of 250,000 RMB for the Robot Taxi, they are preliminary calculations and have not yet reached the level of detailed calculations. We look forward to seeing how they will actually land.
This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.