Author: Yungo
Didi’s autonomous driving, which has always been relatively low-key, has become high-profile since April 2021.
On April 8th, Didi Autonomous Driving Company released a video of autonomous driving continuously for 5 hours without handover. The video recorded Didi’s autonomous driving demonstration in Jiading District, Shanghai. It includes scenes of daytime, evening, and night, and passes through commercial areas, factory areas, office areas, residential areas, publicly disclosing the actual situation of Didi’s autonomous driving demonstration application process.
On April 19, Didi Autonomous Driving announced the new Didi Gemini hardware platform, which is comprehensively upgraded in multiple dimensions such as electronic components, entire vehicle and user experience. It can achieve multi-layer redundancy in four aspects: core high-performance sensors, vehicle-mounted autonomous driving system, remote assistance system, and pre-installed mass-produced vehicle models.
In addition, during this year’s Shanghai Auto Show, Didi also signed a strategic cooperation agreement with Volvo, marking a new stage in the cooperation between the two parties in the field of autonomous driving.
Have to Enter the Game
Didi’s autonomous driving application scenario is mainly Robo-taxi, which is a taxi service that uses autonomous driving technology to replace human drivers.
With the continuous growth of China’s urban population and total road mileage, urban travel demand has also increased year by year. At the same time, the advantages of public transportation and shared travel are becoming increasingly obvious in the face of factors such as urban license plate restrictions, purchase restrictions, and traffic congestion. In addition, as labor costs continue to rise year by year, the cost burden of enterprises is becoming heavier. The emergence of autonomous driving taxis is expected to solve the pain points of the travel service industry. Robo-taxi is gradually recognized as the best scenario for the application of autonomous driving technology.
At present, the commercial application of unmanned taxi services is developing rapidly in various countries. Several countries have allowed autonomous driving passenger-carrying vehicles to operate without safety personnel, and the operation scope of Robotaxi is gradually expanding.
In addition to Didi’s focus on autonomous driving taxis in the domestic market, companies such as Baidu and Pony.ai are also involved. In 2020, Baidu’s unmanned taxi service landed in Beijing and officially opened its autonomous driving taxi service. In April 2020, Pony.ai’s autonomous driving cars in China and the United States exceeded 100, including more than 40 for Robotaxi operations in Guangzhou, and 60 for road testing.
In addition, Didi’s ride-hailing business has reached its ceiling and continues to seek breakthroughs in business directions such as takeout, errand, and community group buying. Combining autonomous driving taxis with ride-hailing business is undoubtedly a highly imaginative prospect for Didi.## Better Financing Results
In 2016, DiDi set up an autonomous driving R&D team and started to focus on high-precision mapping, perception, behavior prediction, planning, and control.
In March 2017, DiDi established its research institute in Silicon Valley, USA for the development of autonomous driving and big data technology. In 2018, DiDi obtained a license for autonomous driving testing in California, USA.
In 2019, DiDi established an independent company for autonomous driving.
In 2020, DiDi’s autonomous driving company received more than 500 million USD in financing. In March 2021, DiDi’s autonomous driving received another 300 million USD financing led by IDG Capital after completing the financing of 300 million USD, and started a new round of financing.
In just one year, DiDi’s autonomous driving have received more than 800 million USD in financing.
In terms of technological development, DiDi has more than 100 autonomous driving vehicles both domestically and internationally, and has obtained road testing licenses in Shanghai, Beijing, Suzhou, Hefei, California, USA and other places.
In June 2020, DiDi passed the CCTV News broadcast of autonomous driving passenger-carrying testing, which won great success in publicity but was not smooth in technology. The vehicle took about two minutes to start during the launch stage, and there were problems such as vehicle deviation and display overlap during the testing process. Moreover, due to heavy rain that day, the cameras and radar were disturbed by rainwater, and the driver had to take over the steering wheel several times.
In the second round of testing, the autonomous driving vehicle had to request remote assistance to change lanes and leave because a tricycle was stopping ahead. From this round of testing, it can be seen that DiDi’s autonomous driving technology can only operate in simple scenarios, and still highly depends on the driver’s operation in slightly more complex scenarios.
Overall, DiDi’s autonomous driving technology is still in its early stages.
More embarrassingly, industry insiders have stated that DiDi’s autonomous driving performance is far less impressive than its financing results.
Data and Operation Advantages
Compared with other companies, DiDi has been deeply involved in the ride-hailing industry for many years, and has its own unique advantages in developing autonomous driving technology.
Data advantages: DiDi’s ride-hailing platform can provide data support for DiDi’s autonomous driving.
In the real environment of autonomous driving testing, the more complex and varied the testing scenarios are, the more perfect the algorithms and models become. Continuous iteration through technology can achieve the highest level of autonomous driving. Faced with complicated road conditions and vehicle driving situations, the testing process of autonomous driving requires massive data support. DiDi relies on the platform’s daily millions of orders, the driving trajectories of drivers, and the monitoring devices installed on vehicles to collect 100 billion kilometers of scenarios annually, which provides sufficient scene sources to DiDi’s autonomous driving.
The biggest challenge for autonomous driving is a large number of extreme scenarios that cannot be handled by rule-based or model-based methods, and it is necessary to solve the rare but critical tail-end scenario challenges beyond 90% of common scenarios in real-world testing. Didi’s advantage is based on real and simulated data to generate operational, safety, and technical maps for its autonomous driving system.
Operational advantage: The ultimate goal of autonomous driving is to serve passengers. Years of operational experience can help Didi explore a better business model within limited capacity and time. Leveraging Didi’s massive operational network, as technology and policies gradually develop, Didi can achieve the goal of mixed dispatching based on ride-hailing orders, which determines dispatch patterns based on road conditions, weather, and routes, where ordinary vehicles and autonomous driving vehicles can provide services to users at the same time and gradually provide autonomous driving services to more people.
Challenges that Cannot Be Ignored
The biggest pressure is cost pressure. Autonomous driving is a long-term and continuous process of burning money, including hardware sensors, software, labor, testing costs, etc. Taking Didi’s autonomous driving demonstration car shown on CCTV as an example, based on Volvo XC60, it is equipped with nearly 20 sensors, including two 16-line lidars, one 64-line lidar, seven cameras, and millimeter-wave radar and ultrasonic radar. Didi COO Meng Xing mentioned that the cost of its autonomous driving vehicles is over RMB 1 million. To invest 1 million autonomous driving vehicles in 2023, Didi needs to invest huge funds of RMB 1 trillion.
In addition to hardware costs, labor costs are also a significant expense. Currently, Didi’s autonomous driving team has over 200 people, and its annual labor cost expenditure is over RMB 100 million. Additionally, Didi’s autonomous driving ride-hailing vehicles need to be equipped with a safety officer and a tester. The cost of cultivating a safety officer is also high, which requires 1-2 months of training with an admission rate of only 1%.
Although Didi’s autonomous driving has raised 800 million US dollars shortly after its independence, compared with the “money shredder” demand in the autonomous driving industry, the high expenditure cost undoubtedly brings tremendous pressure to Didi.
Technology is the fundamental reason that determines how far Didi’s autonomous driving will go.As the application of autonomous driving gradually expands, the safety issues caused by immature technology are gradually exposed. In the current hot Robo-taxi application, most regions still hold cautious attitudes towards it. If serious traffic accidents involving unmanned taxis occur, the government may suspend Robo-taxi services and become more cautious about autonomous driving technology.
In the short term, it is very difficult to achieve Level 4 autonomous driving on completely open roads. What we see now is either low-speed or in simple scenarios, and it is impossible to achieve tests without a safety steward in the short term.
In addition, autonomous driving needs to get rid of fixed-route tracking. Didi’s operation in Jiading, Shanghai is also on a fixed route. Currently, Robo-taxi operations still feature short-distance with fixed stations. In the future, people will certainly not be satisfied with unmanned taxis that only travel 10 kilometers and have a single station.
Unsatisfactory experience is also a major hurdle to promotion. Passengers who have experienced autonomous driving taxis gave feedback: “As an experienced driver, I think current autonomous taxis are a bit stupid.”
Didi’s autonomous driving pressure not only comes from technology companies such as Baidu, Pony.ai and Auto X, but also from Gaode Map. In Guangzhou and Shanghai, Gaode Map has respectively cooperated with autonomous travel company Pony.ai and autonomous driving company Auto X.
Unlike Didi, Gaode Map is a connection platform between passengers and drivers, aggregating more than ten taxi-hailing platforms for passengers to choose freely. Autonomous driving ride-hailing is also the same. Gaode Map itself does not have autonomous driving technology, it cooperates with autonomous driving companies to provide users with free autonomous driving ride-hailing services.
Affected by both technology and capital, the commercialization of autonomous driving is bound to be difficult and requires long-term technology accumulation. In this process, profit is even more difficult to come by. With technological iteration, the hardware costs of autonomous driving components, sensors, and equipment will decrease, and autonomous driving may usher in a new era.
However, currently, Didi’s autonomous driving is more slogan than reality.
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.