Who is positioned to be the leader in autonomous vehicles?
Google? Tesla? Is no one?
Since Google’s Self-Driving Car caused its first accident in February — possibly karmic outfall after its chief executive John Krafcik dismissed the viability of his competitors’ strategies — Tesla and its founder Elon Musk have accelerated their autonomous vehicle (AV) ambitions.
On March 31, the billionaire unveiled the Model 3, billed as Tesla’s first “affordable” electric car and seemingly an overnight success attracting 325,000 preorders and counting. Musk’s post-conference social media activity only fueled existing speculation: “Will the Tesla Model 3 be the first truly self-driving car?” (The aptly titled New Yorker piece concisely summarizes key technological differences between Google’s and Tesla’s AV strategies, though it doesn’t address what “truly” might mean here.)
Then on April 9, Tesla announced its autopilot systems have collectively logged 47 million miles in the past six months, or about 1.8 million miles per week. As of March 2016, Google has only logged 1.5 million AV-miles… since 2009. To begin to determine who maintains the competitive edge, we need to look more closely at the math and issue a few qualifying statements.
Tesla’s modular-at-scale approach to AV technology has enabled it to amass volumes of data. By one estimate, Tesla has 70,000 vehicles equipped with its autopilot system, mainly activated on highways (though restricted from activation in some cases, given controversies related to their drivers ignoring safety advice from the computer systems).
In comparison, Google’s fully autonomous approach has been deployed more incrementally. It operates a mere 50 prototypes and only just announced its fourth city for AV testing — but city miles require much more sophistication to navigate than highway miles. Furthermore, Google also logs more than 20 million simulated AV-miles per week in its labs, no trivial matter given Google DeepMind’s recent advances in artificial intelligence.
How do we assess these performances? Are AV-miles in cities truly more valuable than on highways, and how should we weigh simulated miles?
In a new report, RAND Corporation puts this mileage into context using simple mathematical assumptions. In the U.S., the automobile fatality rate is about 1.09 per 100 million miles, implying that AV systems must drive 8.8 billion miles to demonstrate statistically significant safety. It concludes:
With a fleet of 100 [AVs] being test-driven 24 hours a day, 365 days a year at an average speed of 25 miles per hour, this would take about 400 years […] an impossible proposition if the aim is to demonstrate performance prior to releasing them on roads.
(Note that RAND Corporation does not differentiate between city and highway miles, which would be interesting to breakdown given their presumed differences in mileage and injury/fatality rates.)
Overall, these findings support the need for more virtual simulation and by more players — even at its current rate of simulated AV-mileage, Google would require nearly a decade of testing to demonstrate reliability. For now, the analysis suggests that we may have to wait far longer than Musk predicts to see the days of fully autonomous vehicles en masse on city roads.
