Tired of waiting, self-driving car makers are doing their own safety tests

There’s a new robocop regulating self-driving cars—and it’s a chipmaker. 

Several driverless car companies, including those that test in Canada, have announced their own, self-determined safety standards for their technology, seemingly superseding regulators that are supposed to set the rules of the road. 

The latest came Tuesday in a major announcement by semiconductor behemoth Nvidia. CEO Jensen Huang claims he knew from the moment he saw groundbreaking work in 2012 by University of Toronto’s Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton that helping autonomous vehicles drive safely would be the chipmaker’s “destiny.”

It’s a predictable turn of events for safety researchers like Francesco Biondi. The University of Windsor associate professor has been dubious of tech companies’ past attempts to self-regulate, noting that auto companies have resisted standardizing their gear-shift designs for many decades, despite concerns those designs could lead to accidents. 

The issue is no longer strictly academic, as automakers are pushing Washington to speed up deployment of autonomous vehicles. Will the U.S. government do so by beefing up oversight teams? Or by removing guardrails?

In Canada, some regulators have been hawkish, with B.C. banning highly autonomous driving systems. Ontario has a consultation underway for new rules for autonomous trucking. Transport Canada has invested in an in-house safety simulator and issued recalls on some advanced driver-assistance systems. The department did not provide responses to questions by deadline.  

Still, anticipating regulatory inertia amid a breakneck race to market, some companies are coming up with their own safety standards, with Nvidia emerging as the default clearing house of data. It says its Halos system, to be rolled out by General Motors, makes it the first company to assess every one of a vehicle’s seven million lines of code for safety. Toronto-based Waabi and U.K.-based Wayve, which recently established a research and development centre in Vancouver, are both presenting at Nvidia’s conference in San Jose this week. 

Gatik, which operates driverless trucks for Loblaw and is eyeing an expansion in Canada, gave safety auditors full access to its system to certify it, based on standards set by the non-profit Underwriters Laboratories, and the U.S. National Highway Traffic Safety Administration’s Voluntary Safety Self-Assessment. In an interview with The Logic, CEO Gautam Narang described the approach as unique, noting that his company, too, is working with Nvidia. 

What’s different about the chips? “Everything is automotive grade,” he said, “with a backup, redundant system already in place.” The chips have the computing power needed to simultaneously analyze data at a large scale from multiple sensors, including cameras, radar systems and lidar trackers, he said. 

Waabi, meanwhile, unveiled a methodology in mid-March for evaluating self-driving simulators it hopes will set the “the new safety standard for the AV industry,” saying its own system achieved a 99.7 per cent “realism score.” 

“When we launch driverless by the end of the year, that product will be much safer than a human,” CEO Raquel Urtasun told The Logic at the end of January. She has called for all AV developers to “publicly demonstrate” the reliability of their simulators. 

Tesla, for its part, voluntarily releases data, which for many years have demonstrated that vehicles using Autopilot crash far less than human drivers. 

Biondi cautioned, however, that these types of audits by for-profit firms are no replacement for scrutiny by independent groups like the Insurance Institute for Highway Safety, a  Virginia-based non-profit; and the U.K. charity New Car Assessment Program, which promotes adoption of United Nations crash test standards. Such oversight is needed, he said, because the human drivers who test autonomous vehicles, or get behind the wheel of consumer driver assistance products, may not have the attention span to stay alert when autonomous driving technology is active. He noted that researchers have been working for decades on how to keep even highly trained pilots focused while autopilot is engaged. 

Despite the growing body of self-reported data, Biondi worries about user error as automated driving goes mainstream, and calls for more driver education on how to use self-driving vehicles. Otherwise, he said, we are “handing a toddler a gun and saying, ‘well, be careful.’” 

The web version of this story was updated to clarify details of Waabi’s safety evaluation systems.

Read Shift—The Logic’s authoritative weekly newsletter on automotive technology industry news—for more; and if you know someone who should be reading it, they can sign up here.