Autonomous truck developer Embark pats itself on the back
January 31, 2019
Embark, the autonomous truck startup, has released what it calls the Embark 2018 Disengagement Report. This is not a public breakup between Embark and its girlfriend. It’s about how well the company’s effort to make trucks drive themselves is going. Apparently it’s going well. Embark is hooting and pounding its chest like a victorious Tarzan. Okay, maybe not hooting.
Though Embark trucks drive themselves on the highway, safety drivers are still required. When a driver overrides all those self-driving electronics to take control, for whatever reason, that event is called a disengagement. In self-driving vehicle testing, success is often measured in disengagements per mile driven; the fewer the better.
Embark reports that in the fourth quarter of 2018, its trucks averaged 1,392 miles between disengagements. Embark doesn’t say how many trucks were involved, but a handful of Embark’s modified Peterbilts have been moving freight for real-life shippers – Electrolux for one – between Phoenix, Ariz., and Los Angeles, on I-10 for some time.
At approximately 375 miles between those cities, Embark trucks average almost four trips from one to the other before a driver had to take over – a disengagement. That must be a good score because Embark issued the report voluntarily. No one made them do it. In fact, they couldn’t wait.
“We strongly believe that transparent, consistent, and public testing data published by each automated vehicle developer is an important first step to building public trust around these systems,” the company notes in the report.
That’s typical PR-speak, more commonly known as BS. The reason they made their results public is that they make Embark look good. Embark’s results appear to be as good as those of a leading self-driving car developer, GM Cruise (GM bought San Francisco-based tech startup Cruise in 2016).
See, California requires some companies testing vehicles to report disengagement results, though as noted Embark is not among them. For 2017, the latest results available, there were 20 testers. Waymo, once known as the Google Self-Driving Car Project, filed the best results with one disengagement for every 5,127 miles driven.
GMCruise came in second with one disengagement every 1,254 miles driven, a score close to, though not quite as good as, Embark’s 1,392. That’s excellent compared to, say, Nissan, at 208 miles per disengagement or Mercedes with one disengagement every one-and-a-quarter miles.
Disengagement reports may not help a company’s public image – unless you happen to be Waymo or GM Cruise.
But does the Embark score really compare to GM Cruise?
Consider that Embark operates pretty much on I-10, a controlled-access, multilane highway, part of the original interstate System. It was engineered for safe, high-speed driving. It’s no coincidence that the company’s well-publicized coast-to-coast test drive just one year ago was between Los Angeles and Jacksonville, Fla. – the length of I-10. Embarks software is learning the highway in detail.
GM Cruise, on the other hand, tests in San Francisco on crowded, hilly, often narrow streets with plenty of cars, trucks, buses, people, bicycles, and all manner of big-city surprises. So 1,254 miles between disengagements in San Francisco is simply not comparable to Embark’s 1,392 miles on I-10.
That is not to say Embark’s score is bad. Since other autonomous truck testers don’t issue similar reports, there’s no way to know. Embark’s 1,392 miles between disengagements may well be stellar.
But forgive me for gagging on Embark’s stated reason for releasing those statistics. Maybe Embark really does believe in the benefits of public disclosure. But you can bet the barn they wouldn’t believe all that strongly if their results were closer to Mercedes than GM Cruise.