Left hand turns are tough, and unprotected lefts — no traffic light or stop sign to guide the way — are even harder. This holds true for both human and robot drivers; even the most skilled self-driving car struggles to make a seamless left turn.
Cruise, the self-driving division of General Motors, knows this to be true. In a video just published, the company calls unprotected left turns “one of the most difficult maneuvers” a self-driving car can perform. That’s why the company has focused on perfecting these turns. Cruise’s self-driving cars routinely execute 1,400 unprotected lefts every day in the complex environment of San Francisco, the company says.
“In an unpredictable driving environment like SF, no two unprotected left-turns are alike,” Kyle Vogt, Cruise’s president and chief technology officer, said in a statement. “By safely executing 1,400 regularly, we generate enough data for our engineers to analyze and incorporate learnings into code they develop for other difficult maneuvers.”
Weird flex? Not really. Cruise is likely feeling a bit cocky after massive, back-to-back investments from SoftBank and Honda, leaving the company with a post-money valuation of $19 billion. This may explain why Cruise feels comfortable strutting out this data point, while other self-driving companies are trying to rein in expectations about when their cars will hit the streets.
Why are left turns so difficult? It’s because they require self-driving cars to be both cautious and assertive at the same time. The vehicle needs to nudge its front end into oncoming traffic while its sensors scan for a gap between cars. Human drivers can communicate nonverbally with other drivers using hand gestures or eye movements; robot cars lack those advantages.
“Unprotected lefts are one of the trickiest things you can do in driving,” Nathaniel Fairfield, a software engineer who leads the behavior team at Waymo, told Popular Science earlier this year.
Waymo, which is one of Cruise’s main rivals, is all too familiar with the challenge of unprotected left turns. During a recent test ride, a reporter with The Washington Post reported that “left turns can be painfully slow” when turning onto a major traffic artery. The Information reported a similar problem in 2017. (I experienced an unprotected left turn in a Waymo vehicle last November in Arizona that was neither painfully slow nor seamless. It was totally fine in my opinion.)
In a separate article about Cruise’s bumpy ride to autonomy, The Information noted that the company had set out to tackle the challenge of left turns using machine learning:
Cruise has developed an algorithm that can figure out how far it can “creep” into the middle of many intersections before trying to make a left turn. (Completing the turn successfully is another matter.) Still, that’s just scratching the surface of what a machine should be able to handle in order to replace humans.
Cruise has said it intends to launch a commercial ride-hailing service in San Francisco in 2019.