Weekly Brief: Uber’s Driverless Tech Jaywalks Into Terrible Week

Commentary By Andrew Tolve

---

11th November 2019

Remember that self-driving Uber vehicle that tragically struck and killed a pedestrian in Tempe, Arizona, back in 2018?

Fast forward twenty months and the National Transportation Safety Board has concluded its investigation into the accident. The results don’t look pretty for Uber. It turns out that Uber’s self-driving car didn’t recognize the pedestrian because it had been programmed to only recognize pedestrians if they were crossing a street in a crosswalk. A jaywalker on a dimly lit road at night was thus as confounding to the Uber self-driving car as a ghost would have been to a human driver.

Unlike ghosts, jaywalkers are a normal part of urban driving environments. People jaywalk all the time, often with smartphones in hand and eyes trained on their screens, making them an especially dangerous hazard for drivers. Yet, for uncertain reasons probably involving a mix of time, money and complexity, Uber engineers left jaywalkers out of the software.

Equally baffling is the fact that Uber engineers created a one-second delay between when the car recognized an object that posed a threat for a collision and when the vehicle took action on that threat. The idea was to avoid a constant barrage of harsh braking every time the vehicle detected something that it was uncertain about. Instead, the software would take a second to determine the nature of the detected hazard and what response was necessary. Fine in theory but, in practice, that second could be the difference between avoiding a crash and killing a pedestrian.

Investigators found the car spent four seconds trying to figure out what Elaine Herzberg, the pedestrian, was before finally deciding she was a bicycle with more than a second before impact. That’s when the one-second delay kicked in, during which time the safety driver was watching a video on her smartphone. Uber’s vehicle did finally initiate braking on its own but only with just 0.2 seconds left before impact – far too late to avoid a fatal collision.

As if things couldn’t get worse for Uber last week, an independent expert then proceeded to announce that Uber is still using Waymo’s software. This comes after a highly publicized cyber theft lawsuit that Waymo filed against Uber in 2017, claiming that a former Waymo engineer, Anthony Levandowski, stole Waymo software that then became foundational to Uber’s self-driving car program. That lawsuit ended in a settlement with Waymo receiving $245M in Uber shares in 2018 plus a good faith promise on Uber’s part to scrub any existing Waymo software from the Uber platform.

It now seems that Uber has failed in that latter respect and will thus have to pay Waymo a significant licensing fee if it wishes to continue operating in its present form. Alternatively, Uber will have to drop the software and develop it in-house, or acquire a third party company in the autonomous tech space. No matter the option, it will cost Uber mightily and “could limit or delay our production of autonomous vehicle technologies,” Uber said on a call with investors.

In other words, one of the perceived front runners in the self-driving-car space is on the ropes. Oh, and if that wasn’t enough, Uber stocks fell last week to an all-time low. Talk about a no good very bad week.