Reality and struggle behind Uber selling its autonomous driving business

Uber has been working on autonomous driving for these years

  More than a year ago, Uber Advanced Technologies Group (Uber ATG), the autonomous vehicle division of Uber, was valued at US$7.25 billion after receiving US$1 billion investment from the Vision Fund of Toyota, DENSO and SoftBank. However, time has passed and Uber is negotiating to sell its self-driving car unit ATG to self-driving car startup Aurora. Aurora is a three-year-old self-driving start-up company. The core members of the team are from Google, Tesla, Uber and other big companies, and it enjoys the title of self-driving “dream team”.
  Once upon a time, Uber’s autonomous driving department could be regarded as a potential competitor to Google Waymo, but now it is facing the situation of being sold. What happened behind it? At present, an Uber spokesperson declined to comment on this information, on the grounds that the company does not comment on such information. Whether the negotiation will succeed is worth paying attention to. If it succeeds, Aurora will triple its current size, and Uber can also get rid of a long-term expensive expense.
  On a cold winter night in December 2008, Travis Kalanick and Gretel Kemp couldn’t get a taxi in Paris, so they came up with the idea of ​​using mobile phones to get taxis. In March 2009, they developed a smart phone APP, users only need to use the mobile phone to enjoy the ride service.
  On July 5th of the following year, the world’s first Uber user made an appointment to complete a trip across San Francisco. Uber then began to grow wildly, and in April 2014 it has opened services in 100 cities around the world. In December 2015, five and a half years after the first trip was born, Uber completed 1 billion trips. Subsequently, Uber was officially listed on the New York Stock Exchange in May 2019, with a valuation of 75 billion US dollars.
  It can be said that Uber has been subject to controversy from its establishment to its listing. First, it was resisted in various countries, then it grew brutally, and it suffered huge losses after its listing. It has always been on the cusp of the storm. Uber has been trying to diversify its development. In addition to ride-hailing services, Uber has also opened Uber Eats takeaways and Uber Freight freight services. At present, Uber Eats and Freight businesses have maintained growth of up to 70%, which is much higher than 27% of ride-hailing business and 23.4% of total revenue.
  But if there are lower-cost travel modes in the future, Uber will obviously not ignore it. According to statistics, drivers spend the most in taxi services, accounting for 80% of the total cost per mile. If self-driving cars are adopted, the cost of taxi services will be reduced. By 2040, the global self-driving cars (Level 3 and above) and Driverless taxi services will grow into a 2.5 trillion dollar market, which is why Uber has invested in autonomous driving.
  As early as 2015, Uber took precautions and recruited more than 50 researchers from the Carnegie Mellon University Robotics Institute to set up an autonomous driving team. After that, it spent 680 million US dollars to acquire the autonomous truck company Otto, which was in its heyday. The number of the team once reached 1,500 people.
  In 2015, most autonomous driving companies have not yet started. Uber already has a large amount of cash reserves, a super-large-scale autonomous driving team, and the world’s largest online car giant. It was once considered the only one comparable to Waymo. Of autonomous driving players.

Unit layout of Uber’s self-driving car module

NTSB officials investigate Uber self-driving car that caused accident
Unavoidable safety issues of autonomous driving

  During the five-year race between technology and time, Uber’s self-driving department has not developed smoothly. It has gone through many scandals and has not seen any profits. According to Uber’s 2020 third-quarter financial report, ATG (advanced technology department, self-driving car research is also one of the directions of this department) and other technologies have a net loss of $303 million.
  For example, at the beginning of its establishment, Uber was eager to push self-driving into public lanes, and repeatedly violated regulations without the permission of the California Department of Motor Vehicles. Due to the issue of the license plate of the self-driving car, the California Department of Motor Vehicles was at war. Then in 2018, Uber suffered the world’s first fully autonomous vehicle fatal accident in Arizona, exposing its flaws in autonomous driving technology.
  It was later discovered that the cause of the accident was that the unmanned vehicle itself found a pedestrian crossing the road, but the autonomous driving software system did not take measures to avoid it. Since then, Uber’s autonomous driving tests in Tempe, Pittsburgh and Toronto have ceased. You must know that Uber’s self-driving car is based on the Volvo XC90 SUV. The XC90 is currently recognized as one of the safest cars in the world. This car adopts a modular self-driving software architecture. The entire self-driving system is divided into positioning, perception, prediction, Several modules, such as planning and control, also rely on high-precision maps. It is reasonable to say that the level of science and technology is much more rigorous than some new forces in building cars.
  An internal manager of Uber ATG believes that at the time, ATG’s software engineering team was basically composed of researchers from universities, more like a laboratory, and researchers directly from universities did not have much to commercialize and implement technology. experience.
  After the accident, Uber stopped road testing and started testing on internal test roads. The company also released an autonomous driving visualization tool AVS (Autonomous Visualization System) in 2019, which is mainly used to display the environment during the operation of autonomous driving. Of course, AVS is not only a visualization tool, but also includes complete data processing procedures and data. Interface definition.
  Some analysts pointed out that the possibility of ATG “failure” still comes from the crisis within Uber. Anthony Lewandowski, the former head of Uber’s autonomous driving technology, worked for Google’s parent company Alphabet. In 2018, he was sentenced to 18 months in prison for stealing Google’s autonomous driving trade secrets. Since then, Uber has continuously reported the loss of team members and the resignation of executives, including Don Burnette, the founder of the original Otto. Therefore, it is difficult to take care of the autonomous driving business.
  Moreover, Uber’s key teams still don’t trust each other. Traditionalists from aerospace agencies and government agencies believe that Uber’s self-driving research and development is too radical, and management cautiousness directly slows down the progress of engineers. Even a small test can take months to be approved.

  The self-driving research and development expenditure is too high, and it is difficult to obtain profits in the short term. It has been unable to arouse investors’ interest by burning money subsidies and expanding the market. Two of Uber’s largest investors, SoftBank and Benchmark, have urged the company to reconsider ATG’s strategy and introduce more external investment. Given that the epidemic has severely affected Uber’s core online car-hailing business, it is no longer so easy to burn more money on the autonomous driving business.

AVS visual effect chart
What are the potential buyers

  The company that intends to acquire Uber ATG is called Aurora Innovation. It is a start-up company that has just been established for about three years, but the three core founders are from Google, Tesla, and Uber, and they are all very core technical personnel. Among them, the founder and CEO Chris Urmson (Chris Urmson) is the co-founder and CTO of the former Google self-driving team, laying the foundation of Waymo’s software.
  Chief Product Officer Sterling Anderson (Sterling Anderson) is the former Tesla Autopilot project leader, leading the design and release of Model X. Anderson also caused some legal disputes when he left Tesla. Tesla accused the California Superior Court of “theft of hundreds of GB of classified documents and attempted to poach at least 12 Tesla autopilot engineers.”
  Chief Technology Officer. Drew Bagnell is a machine learning expert. He was one of the founding members of the Robotics Institute of Cameron University and the core backbone of Uber’s autonomous driving team.
  Not only is the team strong, Aurora’s circle of friends is also very strong. LinkedIn founder Reid Hoffman, known as the king of Silicon Valley contacts, is one of its external board members. Companies with in-depth cooperation include car companies such as Volkswagen, Byton, and Hyundai, as well as companies with freight demand such as Amazon. At present, the company has completed a total of 700 million US dollars in financing, investors including Sequoia Capital (Sequoia), T Rowe Price and Amazon, etc., with a valuation of up to 2.5 billion US dollars.
  Aurora’s business model is different from that of general autonomous driving companies. Starting from the L4 scenario, Aurora positions itself as a “full-stack autonomous driving solution provider”, providing auto companies with autonomous driving technology, and will not become a travel operator. Quotient. The technology provided includes maps, perception, planning and control, simulation testing and related hardware and data storage infrastructure.
  As we all know, the technology chain of autonomous driving is very long, and each piece of technology requires a separate team to do it, but a complete autonomous driving requires the seamless cooperation of various technologies.
  Many autonomous driving teams are prone to inefficiency problems when different technologies are coordinated. For example, if the perception team first makes a solution, it must be thrown to the motion planning team to make path decisions accordingly. If the perception team comes up with a new solution For ideas, the sports planning team needs to be re-adapted, which will invisibly slow down the progress of the research. Aurora believes that these are not problems. They have very powerful machine learning capabilities. They can better adapt the interface of two different parts through algorithm design and use machine learning. The company’s official website reads: “We have integrated decades of engineering and scientific experience to create world-class drivers and provide the most powerful and reliable autonomous driving technology.”