A self-driving Uber car in Tempe struck and killed a woman in 2018. Nicole Schaub, Arizona Republic
The self-driving car in last year’s fatal Uber crash in Tempe couldn’t anticipate the actions of jaywalkers and also wasn’t designed to slam on the brakes to reduce the severity of an unavoidable accident, according to details revealed in a new federal report.
Elaine Herzberg, 49, was killed crossing Mill Avenue outside of a crosswalk just south of Curry Road on March 18, 2018, when a self-driving Uber, one of dozens in Arizona at the time, hit her.
Behind the wheel of the Volvo XC90 Uber vehicle was 44-year-old Rafaela Vasquez.
The National Transportation Safety Board released more than 400 pages of documents this week ahead of a Nov. 19 meeting during which the board will discuss the probable cause of the crash.
Also in the new reports were additional details of Vasquez’s training and work history and a recap of her inattentiveness.
Tempe police previously determined from the in-car camera and cellphone data that she was watching an episode of “The Voice” while operating the test vehicle at the time of the crash.
Uber terminated employees nine times leading up to the crash for violating its policies prohibiting using mobile devices while operating the test cars. In nine other instances, it provided additional training for such infractions, according to the new report.
“We regret the March 2018 crash involving one of our self-driving vehicles that took Elaine Herzberg’s life,” Uber said in a statement after the NTSB released the new reports. “In the wake of this tragedy, the team at Uber ATG 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 once issued after the NTSB’s board meeting later this month.”
Already, Yavapai County Attorney Sheila Polk opted not to prosecute Uber after the case was handed to her to avoid a conflict with Maricopa County.
Polk kicked back the possibility of charges against the vehicle operator to Maricopa County after deciding not to go after the company, which pulled its operations from the state.
The Maricopa County Attorney’s Office is still deciding whether it will file any charges against the driver, spokeswoman Jennifer Liewer said Wednesday.
The NTSB previously reported that both the factory-equipped emergency brakes on the Volvo and the autonomous braking added by Uber were shut off, and the new report adds that Uber engineers also did not design the system to make predictions for jaywalking pedestrians.
Uber system didn’t classify Herzberg as a pedestrian
The Uber system controlling the car used radar, cameras and lidar — which is similar to radar but with lasers — to detect objects.
“The detected objects are incorporated into the virtual environment, and the system dynamically updates the vehicle’s motion plan to avoid potential conflicts,” according to the report.
The cars would classify objects as vehicles, pedestrians, bicyclists or “other,” indicating an unknown object.
Once an object was classified, the system would attempt to predict where it was going, or its “goal.”
Previously the NTSB reported that the Uber vehicle never classified the woman as a pedestrian, instead struggling to determine whether she was another car, unknown object or bike.
The new report explains why the car had that problem.
“The system design did not include a consideration for jaywalking pedestrians,” the new report said.
Also, “pedestrians outside a vicinity of a crosswalk are also not assigned an explicit goal. However, they may be predicted a trajectory based on the observed velocities, when continually detected as a pedestrian.”
The Uber system never classified Herzberg as a pedestrian, and even if it did, would have needed to continually classify her that way to predict where she was going.
WHAT HAPPENED: Who was at fault in fatal Uber crash? Here’s the whole story
Uber wouldn’t brake
The Uber vehicles were designed to make minor changes in their trajectory to avoid obstacles. But when hard braking was required, the cars initiated an “action suppression” mode for 1 second.
During this time, the vehicle would verify the nature of the hazard and calculate an alternate path or allow the operator to take over. The Uber system suppressed braking during this period.
This is what happened in the Herzberg accident, and once the car came out of the “action suppression” mode, it was too late to brake hard and avoid the crash.
So the car braked gradually and notified the driver.
The report says Uber has since changed this programming so that its cars will brake hard to at least mitigate an unavoidable crash.
Struggling to classify the hazard
The new report shows the Uber system constantly reclassified Herzberg in the nearly 6 seconds leading to the crash and explains why that was a problem.
When it would reclassify Herzberg, it would have to start over predicting where she was going without taking into account her previous position.
Herzberg was first detected when she was in the far left turn lanes of Mill Avenue and the Uber was traveling in the right lane.
For much of the time the car detected her, it predicted she was a car or bike heading northbound, the same direction as the Uber, according to the report.
The new report offers the following timeline:
- At 5.6 seconds before impact, the car’s radar detected Herzberg, classified her as a car and estimated her speed.
- At 5.2 seconds the lidar system detected her and classified her as an unknown object with a static path. Unknown objects were not considered obstacles unless they were in the path of the car, which she wasn’t at this time.
- At 4.2 seconds before impact the lidar again detected her but reclassified her as a vehicle, predicting the “vehicle” would continue to travel in the same direction the Uber was heading because it was in the left travel lane.
- In the next seconds the lidar reclassified her several times as either “vehicle” or “other.”
- At 2.6 seconds before impact, lidar classified the woman as a bicycle. The new classification restarted the tracking history, and the car predicted her path as “static.”
- At 2.5 seconds it again classified her as a bike that, like the vehicle classification earlier, would continue to travel in the same direction as the Uber.
- At 1.5 seconds before impact, the lidar reclassified her again to an unknown object, but now being partially in the Uber’s travel lane. The Uber planned to steer right to avoid the object.
- At 1.2 seconds the lidar classified her as a bike moving into the path of the Uber, and determined the prior plan to steer around her as no longer possible.
At this point, the Uber went into its 1-second “action suppression” mode.
It came out of the suppression mode 0.2 seconds before impact. The car could not safely brake and swerve to avoid impact, so it began a “controlled slowdown” and gave the operator an auditory alert. The Uber was traveling 39 mph at this instant.
The vehicle operator hit the brakes 0.7 seconds after the impact.
Some systems turned off
The 2017 Volvo XC90 came factory-equipped with a forward-collision warning system and an emergency-braking system. On top of that, Uber installed its own auto-driving system.
The Volvo systems were turned off when the Uber vehicle was operating in autonomous mode. This fact was previously reported, but the new expanded report explains why.
It said a car simultaneously using the radar used by Volvo and the radar used by Uber would have had a high likelihood of misinterpreting the signals because they used the same frequencies.
Also, the brakes on the car were not designed to prioritize commands from one system over the other.
QUESTIONS REMAIN: Will operator in fatal Tempe crash be charged?
Several other crashes
The new report says that from September 2016 to the time of the fatal crash, Uber had 37 other incidents involving test vehicles in Tempe. Nearly all were caused by other drivers.
Twenty-five times, other drivers rear-ended the Ubers, and eight times other drivers sideswiped them.
In one incident, an Uber hit a bent bike-lane bollard that was partially in its lane. On another occasion, the Uber operator took control to avoid an oncoming car that entered the Uber’s lane. The operator struck a parked car in that instance.
In two incidents, pedestrians damaged Ubers that were stopped, according to the report.
Read or Share this story: https://www.azcentral.com/story/money/business/tech/2019/11/06/new-details-emerge-fatal-tempe-self-driving-uber-crash-pedestrian-elaine-herzberg/2508011001/