Tesla's semi-autonomous Autopilot system is collecting huge amounts of data, and that could help it develop a Google-like advantage.
The semi-autonomous driving system, launched in its Tesla-designed form in October 2016, is designed to one day support fully autonomous driving as the system software develops inside Tesla's labs. During the company's first-quarter 2020 earnings call last week, CEO Elon Musk explained how a giant fleet of vehicles on the roads, always training data, brings a big benefit to Tesla – one that could help it outpace the competition.
This is difficult to fully appreciate. It’s the reason, I say, that it's very difficult to have a search engine that competes with Google, because everyone is training Google all the time with their searches. So, when you search for something and you click on a link, you're training Google every time you do that. It's very difficult for any new search engine to compete on that basis.
It's a plan that could help Tesla develop self-driving faster, while also potentially extending its lead once it gets there. MIT researcher Lex Fridman estimates that there are over 800,000 Tesla vehicles equipped with the necessary cameras to support Tesla's full autonomy system. Even those vehicles that aren't using the semi-autonomous features that exist today, which can perform maneuvers like exiting the highway at the correct turning, are still helping the system develop. A "shadow mode" helps collect data even during non-autonomous driving.
On the Autopilot side, Tesla has rapidly built up a large collection of data. In July 2016, Musk wrote that the company could receive regulatory approval once it reached somewhere around six billion miles driven using its autonomous systems. It reached 44 million in April 2016, 222 million in October 2016, one billion in November 2018, and three billion in April 2020.
Musk explained during the earnings call that the Autopilot confirmations, where the driver is asked to confirm that the car should indeed turn off at the next exit, are helping to label Tesla's data and reduce the amount of human-led data labeling required.
All of those confirmations are training our neural net. Essentially, the driver, when driving and taking action is effectively labeling reality as they drive and making the neural net better and better. This is an advantage that no one else has, and we're quite literally orders of magnitude more than everyone else combined.
Tesla may be set to put these abilities to the test soon. It's planning to launch an Uber-like autonomous taxi service, which will enable vehicle owners to let their vehicle join a fleet of Teslas in an area and earn money for the owner. The launch date is unclear, but Musk has suggested the software could be finished from a functional standpoint by next year. Ahead of this launch, Musk has claimed the firm's "full self-driving" package, which grants access to these advanced autonomous features and includes a computer upgrade, will also rise in price.
The Inverse analysis – Tesla's built-in advantage could help it outpace the likes of Waymo, which in January announced it has collected over 20 million miles of fully-autonomous driving through initiatives like its taxi service in Phoenix, Arizona. Unlike Tesla, Waymo does not have a fleet of vehicles collecting data on the roads, but also unlike Tesla, Waymo's cars are running from A to B autonomously instead of in limited situations.
Perhaps the biggest barrier to a Google-like advantage is how Google is free, but the barrier for entry to buying a Tesla still starts around $35,000. The economics could change with the launch of a taxi service that grants access for a small fee per ride, perhaps indicating one of the big advantages of Tesla running its own taxi fleet.