Author: Michelin

The terms of computing power competition, mass production of LiDAR on vehicles, and the booming investment market have made 2021’s autonomous driving industry controversial and exceptionally hot. Along with the disputes, the competition of autonomous driving has entered the second half, and the commercialization has become the key factor for determining the victory, not only for the software and hardware technology and algorithms of autonomous driving.

On November 17th, Baidu released its Q3 financial report for 2021, which disclosed that Baidu’s Q3 revenue reached RMB 31.9 billion, a year-on-year increase of 13%, and its net profit attributed to Baidu was RMB 5.09 billion. Of the revenue, RMB 5.2 billion came from core non-advertising revenue, such as cloud computing and Apollo intelligent driving. This means that the autonomous driving business, which was thought to be in a “burning money” mode, is expected to become a core business that contributes to Baidu’s revenue.

The commercialization era of autonomous driving has started.

RoboTaxi’s Order Quantity Ranks First Globally, with over 270,000 Orders in Half a Year

If autonomous driving is like a distant and beautiful paradise, then commercialization of technology and products is the bridge that provides forward advancement. Baidu often uses a sentence to describe their autonomous driving, “Climbing Mount Everest, laying eggs along the way,” meaning commercialization is significant for achieving fully autonomous driving.

The Q3 financial report of Baidu shows the results of “laying eggs along the way”: The core non-advertising revenue of Apollo intelligent driving and cloud computing has increased year-on-year. According to Baidu, RoboTaxi’s order volume was 115,000 in Q3 2021, and if estimating according to this growth trend, RoboTaxi’s cumulative orders in the second half of the year may reach 270,000. It is worth noting that Waymo, another giant in autonomous driving, had an order volume of around 1,000 to 2,000 per week in 2020, and is estimated to have a quarterly order volume of 20,200 to 40,300, and an annual order volume of about 160,000 after the fleet expansion in August 2021.

This also means that Apollo has surpassed Waymo to become the global leader in order volume. Both the familiar L4 level RoboTaxi scene and the reduction in dimension and application of autonomous driving technology on mass-produced cars have contributed to Baidu’s order ranking first globally.

L4-Level Autonomous Driving Landing, Scale is King## Robotaxi: the Key to L4 Level Autonomous Driving

Robotaxi is arguably the most important scene for L4 level autonomous driving, where companies such as Waymo, Cruise, and Baidu’s Apollo are placing their bets on. It is also the most challenging landing scene for autonomous driving, as it requires not only tackling complex real road conditions, but also gradually establishing trust from users regarding unmanned driving mode, and balancing operational costs with revenues.

During the Q2 earnings call, Robin Li, CEO of Baidu, stated that when there are approximately 50 million orders per day, Robotaxi’s operating costs will be one-fifth of what they are now. The key to reducing costs is the scaling effect brought about by the rapid expansion of the fleet and orders.

According to Baidu, the Fast Rabbit platform, which operates during Q3, provided 115,000 travel services. The growing Fast Rabbit is making Baidu’s business model clearer, and the scale brought about by the growth in orders also adds new competitiveness for Baidu in the Robotaxi industry.

In addition to commercial scenarios, the most important autonomous driving scenario for users is probably the intelligent driving functions on their own vehicles. The level of vehicle intelligence has gradually become a deciding factor for users when buying a car.

The demand for intelligent driving functions on mass-produced vehicles has prompted OEM to choose to invest in full-stack self-developed solutions with long cycles, or to partner with autonomous driving technology companies like Baidu to provide intelligent driving solutions.

For example, in October, Baidu Apollo and WM Motor jointly released a mass-produced model based on the W6, which features Baidu’s ANP Navigation and Parking Management System and AVP Autonomous Parking System, as well as the Apollo Moon version vehicle based on the “ANP-Robotaxi” architecture. This architecture enables the symbiosis and sharing of L4 level Robotaxi autonomous driving data and L2+ level auxiliary driving data, as well as the reduction of technology dimensions and the feedback of data. This feedback not only satisfies the needs of Robotaxi unmanned operations, but also accelerates the rapid deployment of autonomous driving.

Technological Accumulation: From Quantitative to Qualitative Change

As more and more automakers, startups, and internet giants join the competition in the field of autonomous driving, Baidu, an “old soldier” who has been laying out the field of autonomous driving for 8 years, tells us with its actions that autonomous driving is a road without shortcuts, and every step on the road requires technological accumulation.

This accumulation comes not only from the improvement of intelligence level, but also from the feedback of data after actual commercial landing.

According to Q3 financial report, Baidu invested RMB 6.2 billion in R&D, an increase of 35% YoY, accounting for nearly 20% of revenue. In the past few years, the R&D investment of Apollo has been above 15%. The most direct manifestation of the heavy R&D mode is the significant improvement in the system’s intelligence level.

In June this year, Baidu released the fifth-generation shared unmanned vehicle, Apollo Moon. The new generation of Apollo Moon not only reduces the cost to one-third of the same type of vehicle in the industry, but also upgrades its smartness through software and hardware improvements, making its intelligence level ten times higher than the previous generation, so that Apollo Moon can handle perception and decision-making with ease. In October this year, Apollo officially launched the “main driving position without safety officer” test in Beijing, which also affirmed Apollo’s automatic driving level by canceling the safety officer at the main driving position.

Another “big problem” in autonomous driving is corner case. The improvement of the intelligence level of a single vehicle can enhance the system’s perception and decision-making ability, but it still becomes helpless in special situations with low probabilities. The solution is to verify the system reliability by a large number of simulations and on-road tests to minimize the probability of corner cases.

In September 2019, when Baidu’s first mass-produced Robotaxi debuted in Changsha, only 45 vehicles served as the “pioneer troops” for testing and operation. Over the past two years, the testing team has expanded by tens of times, with 411 autonomous driving test licenses already acquired, and plans to expand the fleet to 3,000 vehicles by 2023.

The expansion of the fleet size means a greater testing mileage: Apollo’s daily testing mileage of the global fleet has reached 40,000 kilometers, which is equivalent to circling the earth’s equator every day if all these testing mileage are connected together. Up to now, Apollo’s multiple L4 autonomous driving models have accumulated a total testing mileage of 18 million kilometers.

Previously, Waymo was the “benchmark” for the industry in terms of testing mileage. Currently, Waymo’s autonomous driving test fleet tests a mileage of 160,000 miles per week, which is equivalent to about 36,000 kilometers of daily increase mileage. However, the route of hard-balling with Waymo’s L4 autonomous driving is different. Apollo has chosen a more easily achievable and scalable business model. However, it must be admitted that domestic tech companies have innate advantages and efforts in scalability.

The accumulation of these technologies and testing mileage has enabled autonomous driving to move from the laboratory to large-scale production and implementation, step by step expanding the commercial boundaries of Apollo. It has also set a goal of expanding the business to 100 cities with Carrots running fast by 2030.

Finally

Since the concept of autonomous driving appeared, many have dreamed of a future where all vehicles on the road are autonomous. Although this dream seemed like a fantasy in the short term, Apollo’s technology and product landing in batches have also allowed us to see the approaching footsteps of the commercialization of autonomous driving.

At the Q2 2021 financial report meeting, Li Yanhong predicted that Carrots Running Fast would be profitable in “2025”. Judging from the speed of Apollo’s commercialization, this goal is getting closer and closer.

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.