A brief history of the five Silicon Valley autonomous driving families' "land grabbing"

Since the birth of the car, humans have never stopped imagining autonomous driving.

"Humans should be freed from driving." In 1940, industrial designer Norman Bell Geddes envisioned that through automatic driving roads, cars could automatically move according to trajectories and programs within a certain range.

This concept was denied in later research, but in the past decade, autonomous driving has been seen as a key direction to change the way we travel in the future. Countless autonomous driving companies incubated in Silicon Valley, which is the most tolerant of new technologies, have launched a fierce offensive against this pearl that represents the highest level of artificial intelligence.

The industry calls Waymo, Cruise, Argo AI, Aurora and Zoox the "five major families" of autonomous driving in Silicon Valley. They are in a leading position in technology and scale, and are also the targets of capital. Looking back at the development history of these "pioneers", we will find that they all came from a DARPA autonomous driving challenge between 2004 and 2007.

Incubator: DAPRA

Just as the earliest nuclear power was not used to generate electricity but to power large military ships, the rapid development of autonomous driving also comes from the technological needs of the military field.

In 2001, the United States was deeply involved in the war in Afghanistan. In order to promote the research and development of autonomous driving technology in the military field, the U.S. Congress authorized DARPA (Defense Advanced Research Projects Agency) to organize an unmanned vehicle challenge and provide a million-dollar prize to the winning team.

In 2004, the first DARPA Challenge was held in the Mojave Desert in the United States. Although no team completed the route or won the prize, it inadvertently kicked off the rapid development of autonomous driving technology.

The Stanford team's self-driving car Stanley won the championship in 2005

At the second DARPA Challenge in 2005, Sebastian Thrun, director of the Stanford University Artificial Intelligence Laboratory, led his team to build a self-driving car converted from a Hummer and won the championship.

In the same year, Thrun met Google founder Larry Page. The two were Stanford alumni and had the same judgment in the field of artificial intelligence: this would be an important direction for human technological progress.

The three DARPA Challenges have cultivated countless talents in the field of autonomous driving in the United States, and Google's autonomous driving project has become an incubator for founders of autonomous driving companies in Silicon Valley. Looking at the founders and developers of startups, they are all inextricably linked to the DARPA competition and Google's driverless car project.

In 2009, Google launched the self-driving car project, led by Thrun, former director of the Stanford AI Lab. Other researchers included Mike Montemerlo and Chris Urmson. The former was one of the creators of Stanley, the 2005 DARPA champion, and the latter led Carnegie Mellon University to defeat Thrun's Stanford team in 2007.

There is no need to say much about Waymo. It was born out of Google's self-driving project. It became independent in 2016 and became an independent department of Alphabet. It has maintained its leading position in the self-driving field to this day.

Zoox, which has a more radical design concept, has a co-founder, Jesse Levinson, who was a favorite student of Thrun when he was in charge of the Stanford AI Laboratory. He wrote a navigation program for the Stanford team in the 2007 DARPA challenge and won second place.

Chris Urmson, the leader of the championship team that defeated the Stanford team, resigned from his position as CTO of Google's self-driving project in 2013 and co-founded Aurora with the former head of Tesla's Autopilot project and the former head of Uber's self-driving project.

Coincidentally, Bryan Salesky, who was on the same team with Chris in the DARPA challenge, embarked on the same path: he worked on Google's driverless project, and then co-founded Argo AI with Peter Rander, former chief engineer of Uber's advanced business technology department, in 2016.

Also in 2016, Kyle Vogt and Dan Kan co-founded Cruise Automation, which was later acquired by GM for $1 billion. Founder Vogt was a top student at the School of Computer Science at MIT and participated in the 2004 DARPA Autonomous Driving Challenge.

At this point, startups incubated from the DARPA Challenge and Google's autonomous driving project have all entered the autonomous driving track and ushered in a golden period of being favored by capital.

Waymo: Dissuade L5 and embrace commercialization

When Sebastian Thrun led the Google self-driving project, he said that commercialization might be the least important thing. "We are most concerned about the technology, not the plan to bring it to market. This may sound unreasonable, but we all know that if we can solve this problem, we can do whatever we want."

Google has never concealed its confidence in its technology and R&D. Its founder Sergey Brin once made a bold statement in 2012: the number of years it will take to achieve fully autonomous driving can be counted on the fingers of one hand.

Regardless of Brin's over-optimism about fully autonomous driving, there is more than one way to Rome, and in the ideal timetable, players in the field of autonomous driving have already chosen their tracks.

Some companies choose to start by building a car, such as Tesla, which develops autonomous driving technology while building and selling cars, gradually upgrading from L2 assisted driving to L3 and L4 to achieve fully autonomous driving; Zoox, which was born in Silicon Valley, has the same idea and set the goal of "directly producing cars that do not require human driving" at the beginning of its establishment.

Waymo has chosen to focus on the research and development of autonomous driving systems. Patrick Cadariu, head of the automotive supply chain, told the media, "Our core mission is not to produce cars, but to produce the most experienced drivers in the world (i.e. autonomous vehicles). Since some companies are very, very good at producing cars, we are willing to cooperate with them."

In terms of R&D path, Waymo does not believe that it can achieve the iteration from L2 to L4, but chooses to go straight to L4. Waymo believes that L2 and L4 are completely different in terms of logic from system architecture to dealing with emergency scenarios, and there is a natural gap between the design applicable to L3 and below and L4+; on the other hand, L3 self-driving cars have encountered dangerous situations due to human operation many times during testing, and achieving L4 in one step can avoid accidents caused by immature technology during the iteration process.

Compared with other Silicon Valley autonomous driving families, Waymo has an absolute leading position in data collection and technology application. From 2014 to 2019, the total mileage of road tests each year has been far ahead. In 2019, the road tests in California exceeded 1.45 million miles, which is more than half of the total test mileage of all vehicles. After Waymo One went online in 2017, its order volume has exceeded 100,000.

But if you look at it longitudinally, R&D progress is much slower than Waymo expected.

At the end of 2017, Waymo CEO John Krafcik began drafting Waymo's business plan, saying that it would launch a fully driverless taxi service by the end of 2018. However, in the trial operation in Phoenix, there were no drivers behind the steering wheels of hundreds of "driverless cars", but safety engineers were still sitting in the back seats, monitoring the driving status of the vehicles at all times.

Krafcik even "persuaded" the plan many times, and took the initiative to state that the previous plan underestimated the difficulty of autonomous driving, saying that "L5 autonomous driving will not be put into practical use within 10 years."

Funding is another issue. Waymo went straight to L4 and gave up the opportunity to make money through L2 mass production. This is not a problem when investors are enthusiastic about autonomous driving, but don’t forget that autonomous driving research and development is a business that burns a lot of money and has no long-term returns.

As Krafcik puts it, “The field is weighed down by the unfulfilled promise of AI, and the downpour of investor interest has now turned into a drizzle as a result of the failure to deliver breakthroughs.”

Entering 2020, capital's enthusiasm for autonomous driving has rekindled and is tilted towards leading companies.

In March, Waymo announced that it had received US$2.25 billion in its first round of financing, led by Silver Lake Capital, Canada Pension Plan Investment Board and Abu Dhabi Fund Mubadala. Together with another US$750 million investment, Waymo's total fundraising from external investment institutions reached US$3 billion. This was also the first time Waymo received investment from outside Alphabet.

How long can $3 billion last? At present, Waymo spends more than $500 million on salaries for its autonomous driving R&D team each year. If you add fleet expenses and operating costs, the annual expenditure is nearly $1 billion. If Waymo wants to achieve autonomous driving as soon as possible and collect more real-life data, it needs to expand its fleet, and the expenditure will be even higher.

In June, Waymo announced that it had become Volvo's exclusive global partner for Level 4 autonomous driving. Volvo will use Waymo's autonomous driving technology to build autonomous electric taxis, and will use Waymo's autonomous driving technology for its two sub-brands, Polestar and Lynk & Co.

It is not yet known when L4 will be achieved, but Waymo is moving towards a more commercialized path.

Zoox: Two-way cars take off the autonomous driving track

A month ago, American e-commerce giant Amazon acquired Zoox for US$1.2 billion, and the Silicon Valley star company finally withdrew from the autonomous driving track by "selling itself".