Autonomous Car Accidents and Policy Implications – Part 5.2

In the last section of this article, we ended with the media subtly but firmly suggesting that the death of a pedestrian hit by an Uber AV was a failure of technology and of policy.  The AV START Act was postponed (something that might have been good for industry overall given the regulatory climate in DC right now), and Uber had suspended testing.

As the days went on, the bad news kept flowing, and two days after the NTSB arrived to begin investigating, a New York Times article went over a myriad of details which had come out prior to and after the crash. These ranged from how they’d planned on having a publicity event where their CEO was taken around Tempe by a fully functional AV in April (canceled, of course) to reports of Uber drivers falling asleep in their AVs. It also mentioned the difficulties faced by AV operators, like how they had to keep their hands ‘hovering’ over the wheel at all times, and pay attention to road conditions for hours in order to take notes. The notes were supposed to be taken only at stops, but often drivers did it while the vehicle was in motion. They contrasted this to Waymo’s system of having a button on the steering wheel which let drivers record short audio clips rather than having to navigate graphical interfaces and possibly type out notes. In fact, the article had several comparisons to make, always praising Waymo as doing the better job.

On the same day that this article came out, Velodyne (the company that makes the LIDAR arrays almost all of us use) made a public statement that they were positive that their sensors were not to blame. Instead, they said “the problem lies elsewhere”, a statement which suggests it was the fault of the AV system’s software. They later added that they believed that Uber’s decision to eliminate side-LIDAR arrays in favor of more radars and keeping its top-mounted 360 sensor as the only LIDAR on the vehicle meant it was less likely to see pedestrians. Velodyne’s representatives said that it created a 3 meter blind spot around the lower perimeter of the car, meaning that pedestrians and other objects which entered that space from certain angles could fail to be seen by the LIDAR. It’s important to point out that the radars and cameras appear to cover this zone and that it’s certainly an edge case to think of an object which enters the visual range of the car without passing through the LIDAR’s view, as well as being transparent both optically and to radar returns. It’s a tough sensor fusion problem when one sensor returns an object and another doesn’t, but if the object is in a space where the LIDAR cannot see at all, then the algorithm would probably be set to take that into account. More sensors are both more expensive and slow the algorithm down because it needs to process more data. While more LIDAR arrays may have helped in this case, it’s more down to the software than the hardware in my opinion. Teslas have only radar and cameras and they show the ability to stop for pedestrians. Additionally, in this specific case in Tempe, the Uber identified a need to stop with enough time to reduce speed to the point where authorities say it’s likely the pedestrian would have survived.

The rest of the AV industry was distancing itself from Uber, suggesting the inference that Uber was a single bad egg and the rest of the basket was doing just fine. Nvidia, which made some of the hardware that went into Uber’s AV – suspended their own testing voluntarily. Boston, which had suspended testing in the city in order to do safety reviews, let nuTonomy and Optimus Ride begin testing again a little more than a week after the crash. Waymo said their car could have handled it and Lyft said the driver should have been able to stop the crash.

Next time we’ll see how the regulatory environment evolved in Arizona to make Uber’s testing there possible, as well as the fallout for the governor who was instrumental in their dominance in the State.  Governor Ducey and Uber try to throw each other under the autonomous bus in part 3.

2 thoughts on “Autonomous Car Accidents and Policy Implications – Part 5.2

  1. Pingback: Autonomous Car Accidents and Policy Implications – Part 5.3 | RxEvolution - Portfolio of Paul Calhoun

  2. Pingback: Autonomous Car Accidents and Policy Implications – Part 5.4 | RxEvolution - Portfolio of Paul Calhoun

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s