NTSB: Autonomous Uber Vehicles Crashed 37 Times Before Fatal Accident

Matt Posky
by Matt Posky
ntsb autonomous uber vehicles crashed 37 times before fatal accident

The National Transportation Safety Board (NTSB) has disclosed Uber’s autonomous test fleet was involved in 37 crashes over the 18-month period leading up to last year’s fatal accident in Tempe, AZ. Having collected more data than ever, the board plans to meet on November 19th to determine the probable cause and address ongoing safety concerns regarding self-driving vehicles.

Reuters reports that the NTSB plans to issue comprehensive safety recommendations to the industry, as well demand oversight from governmental regulators, in the near future.

Unfortunately, the circumstances surrounding the fatal incident in Arizona are as unique as they are complicated — ditto for most other crashes involving AVs. While Uber’s test mule failed to identify the pedestrian in time, leading to her death, she was also walking her bicycle on a particularly awkward stretch of road. “The system design did not include a consideration for jaywalking pedestrians,” the NTSB said.

The vehicle’s safety driver was also totally engrossed in their phone. Local authorities said the crash would have been avoidable if the backup driver had bother to look up — something that’s tragically obvious from viewing the dash-cam footage. This also helped prosecutors decide Uber would not be criminally liable for the crash. However, the NTSB identified numerous issues with the autonomous systems the company had installed at the time.

From Reuters:

The NTSB reported at least two prior crashes in which Uber test vehicles may not have identified roadway hazards. The NTSB said between September 2016 and March 2018, there were 37 crashes of Uber vehicles in autonomous mode, including 33 that involved another vehicle striking test vehicles.

In one incident, the test vehicle struck a bent bicycle lane post that partially occupied the test vehicle’s lane of travel. In another incident, the operator took control to avoid a rapidly approaching vehicle that entered its lane of travel. The vehicle operator steered away and struck a parked car.

NTSB said Uber conducted simulation of sensor data from the Arizona crash with the revised software and told the agency the new software would have been able to detect the pedestrian 88 meters (289 feet) or 4.5 seconds before impact. The car’s system would have started to brake 4 seconds before impact.

But the car did detect her. It just had trouble recognizing what kind of object she was and failed to make a decision while the safety driver wasn’t paying attention. Uber now applies maximum emergency braking to prevent a similar situation from happening, as indicated above.

Uber’s Advanced Technologies Group didn’t have an operational safety division or dedicated safety manager prior to the crash, let alone an official safety plan. But it’s not legally obligated to have one under current AV laws, so why would it have bothered? Hindsight is 20/20 — not foresight.

Sarah Abboud, a spokeswoman for Uber’s self-driving unit, said the company deeply regretted the crash that killed Elaine Herzberg and has “adopted critical program improvements to further prioritize safety.”

“We deeply value the thoroughness of the NTSB’s investigation into the crash and look forward to reviewing their recommendations.”

[Image: Uber/Volvo]

Join the conversation
3 of 23 comments
  • Sirwired Sirwired on Nov 08, 2019

    Self driving has come a long way, but still has a *long* way to go. Here's a non-Uber example: For some reason Chrome thinks I want to read everything, everywhere, about Tesla. Recently, one of the Nonstop-Tesla-Cheerleader sites posted an article with the title of "Autopilot successfully navigates construction zone lined with cones." (I didn't read the article, but I guess the implication was that it succeeded in not robotically using the lane markings?) If "successfully performed task even the most addle-brained learner's-permit teenager can accomplish with ease" qualifies as news, lets just say that I'm not confident in His Muskiness' plan for "Full Self-Driving" any time soon.

    • Hummer Hummer on Nov 08, 2019

      “ reason Chrome thinks I want to read everything, everywhere” All of the tracking software I see other get bombarded by don’t work on my devices for some reason (not that I mind). I get advertisements for businesses in Ohio or California all the time, haven’t even been to either state in several years. And all of my advertisements come for things I would never actually buy, like Rolls Royces, advertisements for stores I don’t shop at or talk about, advertisements for pressure washing services in Florida is another I constantly see, which is stupid because I would never pay someone else to do that.

  • Don Mynack Don Mynack on Nov 10, 2019

    “The system design did not include a consideration for jaywalking pedestrians,” the NTSB said. It's cool, just not enough story points in the next few sprints for that, will get around to once somebody grabs it off the backlog.

  • Parkave231 On the one hand, I always thought that TriPower was a horrible name for the L3B engine (and LSY, although I don't recall seeing that name applied in practice), especially since one of the three technologies reduces said power (AFM). Of all the historic GM names to bring back...On the other hand, TurboMAX is a horrible replacement name. Turbo-Quad4? (Yes, I know it's not.) FleetFour? (Seemingly target market?) FourReal?
  • SCE to AUX The diesel isn't that compelling compared to the 2.7T, when you consider the 50% fuel cost premium and the need for DEF.But regularly towing 9500 lbs with a 4-cylinder (even a low-stress one like this) seems to be overdoing it. I'd get the 4 for lighter duty, the diesel for medium duty, and one of the 8s for heavy duty.
  • Analoggrotto Over the years GM has shown a keen interest in focusing their attention and development money on large, expensive or specialized vehicles and little to no progress in developing something excellent to complete with such class leaders as : Camry, Telluride, Civic, CR-V, Highlander, Accord, or even ho hum Corolla. And this is the way class division works in the heartland/rustbelt: pretend to care for the common man but cater the public resources to additional security and comfort for the upper echelons of society. GM is Elitist American Communism.
  • Art Vandelay Current Fiesta ST
  • Jeff S Buick Lacrosse and Chevy Montana compact pickup.