Arizona officials on Wednesday released grainy footage of the Uber self-driving car crash that killed a 49-year-old woman.
The video may pose more questions than it answers.
The video stops moments before the impact that killed 49-year-old Elaine Herzberg. It depicts a dark night, a figure that emerges a couple of seconds before the crash, and a seemingly distracted human safety driver, Rafaela Vasquez.
Tempe Police Vehicular Crimes Unit is actively investigating— Tempe Police (@TempePolice) March 21, 2018
the details of this incident that occurred on March 18th. We will provide updated information regarding the investigation once it is available. pic.twitter.com/2dVP72TziQ
At best, the video is inconclusive on what—if anything—could have prevented the crash. At its worst, it brings up dozens of questions that safety officials will have to answer in upcoming months regarding self-driving vehicle testing, regulation, and liability in crashes.
Uber released a statement Monday and Wednesday and said its fleet of self-driving cars in Pittsburgh, Toronto, San Francisco, and Tempe, Arizona would be halted pending the investigation.
Our hearts go out to the victim’s family. We’re fully cooperating with @TempePolice and local authorities as they investigate this incident.— Uber Comms (@Uber_Comms) March 19, 2018
Uber hasn’t commented on what, if anything, its self-driving car’s sensors recorded that night. The company also hasn’t said if all systems were functioning properly, or if it was testing a previously unreported sensor array.
A spokesman for Volvo told Motor Authority on Wednesday that they "are aware of the incident and have seen the disturbing and upsetting video. Our thoughts remain with Elaine's family and friends. Uber is cooperating with local and national authorities and Volvo is assisting in these investigations."
What we can and can't see
Most companies developing self-driving software closely guard how their sensor data is interpreted by computers in the car, and it’s unlikely that Uber would disclose much publicly outside of a court order.
Based on information they’ve publicly shared, and information Volvo has shared about the stock XC90 SUV driven that night, it’s clear that one—or more—safety systems malfunctioned, or were disabled in the crash.
Volvo has made its City Safety suite of advanced safety technology standard on all XC90s since 2016. That suite includes automatic emergency braking with large animal detection. In 2016, the IIHS rated the system as “Superior” and noted that the system avoided a collision with a wall at 25 mph.
At speeds up to 62 mph, Volvo’s City Safety system should help steer a car out of a collision with a large animal, according to the automaker. Reports say the self-driving car by Vasquez was traveling at 38 or 40 mph when it struck Herzberg.
Uber ATG tech
That large animal avoidance system was partially developed and tested in 2015 in Australia to help avoid collisions with kangaroos, according to Volvo. That addition to City Safety made its debut in 2017 in the XC60 and has since been applied to other models.
According to Volvo, the system uses radar and high-definition cameras to detect large animals, pedestrians, and cyclists ahead of the car. Radar should “spot” the object, and highly light-sensitive parallax cameras confirm if the object is in the intended path of the car.
According to Volvo, the system reacts within .05 seconds, compared to 1.2 seconds for a typical human driver. At 40 mph, a human driver would need roughly 80 feet to react to the obstacle and stop the car.
Radar and camera-based systems are hardly new, and many automakers have refined those systems to work in low light. Poor weather conditions can impair camera-based systems; the National Weather Service reported “no significant weather events” for the Phoenix area. Weather records showed 24 percent average humidity that dropped steadily throughout the day, to as low as 7 percent at 5 pm; an average wind speed of 4 mph; a high of 74 degrees and a low of 50 degrees. The sun set around 6:39 pm.
Cadillac Super Cruise LiDAR scan
The combination of mild weather and Volvo’s sensors may not matter, ultimately.
Uber’s fleet of self-driving cars is equipped with multiple sensor arrays that include lidar to mitigate poor visibility, poor weather, or both.
It’s unclear what the lidar sensors saw that night—it’ll likely be a main area of interest for safety investigators—or if the array was functioning properly.
Previous demonstrations by suppliers and manufacturers of lidar, which is an acronym for “light detection and ranging,” have shown off the systems’ capabilities in low light. The lidar sensors only have a rudimentary “view” of the outside world, but most observe basic shapes, size, and closing speed relative to the vehicle, and can scan the car’s surroundings with a 360-degree view.
Without those sensors, Uber relies on their human safety drivers to react to avoid crashes. It’s unclear what Vasquez may have been looking at from the video, but it wasn't the road ahead. Her hands also weren’t placed near the wheel, which is what Uber instructs its safety drivers to do before testing on public roads, according to the New York Times.
According to CNN, Uber requires its safety drivers to train for three weeks, and to take a written test, before they can operate test vehicles on public roads.
The video shows Herzberg crossing the road outside of the area illuminated by a street light and wearing dark clothes. Perhaps even an attentive driver wouldn't have seen her.
Maps of the area show a bus stop near the area where Herzberg was struck, on the west side of North Mill Avenue, and a divided roadway leading toward East Curry Road, the nearest intersection to the crash. A signaled crosswalk at the intersection of Curry Road and Mill Avenue is roughly 210 feet north of the bus stop, which is directly across the road from a park sign where photos showed Herzberg’s crumpled bicycle Sunday night.