The Insurance Institute for Highway Safety (IIHS) does more than crash testing. The group also conducts surveys, studies, and analysis of safety systems new and old. In a recently-published study, the IIHS looked at the Waymo “robo taxi” (self-driving taxis) experiments being conducted in various cities around the nation.
The study’s results found that a Waymo is far less likely to get into a police-reportable accident than is a human. The study was based on publicly-available traffic information taken from police jurisdictions where the robo taxi company is conducting operations. Despite the social media trend of showing all of the Waymo blunders that happen, the reality is that, according to IIHS, a Waymo vehicle is 68 percent less likely to get into a serious accident than is a human-driven vehicle.
The actual number is 68 percent fewer police-reportable accidents per vehicle mile traveled. And, to be fair, the Waymo vehicles are under far stricter reporting requirements than are us regular humans. The comparison was per mile driven, so the 50 million Waymo miles driven versus 222 billion human-driven miles becomes an even playing field. And, we’ll see, is also why it’s Waymo and no other automated driving companies in the study.
Waymo is currently operating in San Francisco, Los Angeles, Phoenix, and Austin in a sort of half-grin portion of what’s often called the “EV smile” region of the U.S. The place where electric vehicles (which most Waymo vehicles are) sell the best. the law does not require people involved in an automobile accident to report it to the police unless significant harm or bodily damage occurred. You or I could get into a fender-bender, trade insurance information, and be on our way without police involvement. For Waymo, any accident, no matter how minor, must be reported. Which means the numbers here were skewed in favor of humans and we still lost. Even before the IIHS researchers removed accidents that would not have likely been reported by human drivers.
What IIHS also said, however, is that this information should have a little bit of salt put on it. The system for monitoring automated vehicles is not exactly uniform. The study’s researchers’ biggest conclusion was not that driverless cars are safer, but that the means for monitoring them is not very clean. So while the result is a big win for autonomous vehicles, it has a caveat.
“The results show that, on a limited scale, these driverless cars are safer than human drivers — who can be impaired or drowsy or suffer lapses in attention,” IIHS President David Harkey said. “However, the present data collection system isn’t good enough to allow continuous monitoring of a large-scale expansion.”
Most of the self-driving vehicle data that’s collected is under the National Highway Traffic Safety Administration’s (NHTSA) Standing General Order on crash reporting for automated driving systems. The SGO was issued in 2021 and is the only national data collection tool for crashes, but it’s limited to being primarily for recall enforcement. It does not collect crash information involving humans in the same way and it doesn’t always delineate between a human-assisted (Leve 2 or 3) system and a fully-automated Level 4.
The other major system is California’s, given that two of the four cities in which the Waymo experiments are happening are there. This system, IIHS says, is not much better than the NHTSA’s.
And for the Tesla fans or those who wonder why this study focused on Waymo: no other driverless car companies or providers include mileage logs. Waymo does. This makes it impossible to know how common vehicle crashes are. If a driverless car gets into two accidents in a year, but there’s no way of knowing how many miles it drove that year, the information is mostly useless. Because Waymo logs and voluntarily provides all of that data, including mileage, it can be analyzed for frequency. This meant that researchers at IIHS had to remove all records sent in from frequency-free records from Cruise, Zoox, and others. Plus incidents where the self-driving system wasn’t engaged. Many Waymo crashes included a human driver who’d disengaged the self-driving system, IIHS said.
The overall results look good for self-driving vehicles. Well, at least the Waymo ones. And it rightly points out that the data isn’t being shared evenly.
This article originally appeared on the AaronOnAutos Substack.
