Automotive > Autonomous Vehicles Blog

Uber ATG's Legacy

by Roger Lanctot | 12月 21, 2020

Uber Technologies has sold its Advanced Technologies Group (ATG) subsidiary to Aurora Innovation putting its fatally flawed autonomous driving endeavors in the rearview mirror. A virtual meeting of the Autotech Council last week highlighted the enduring legacy of Uber ATG and its contribution to the business of developing self-driving cars. That contribution derives from the untimely death of Elaine Herzberg on March 18, 2018, in a fatal encounter with an Uber self-driving car.

The crash, for which the safety driver has been charged with negligent homicide, was a result of a variety of mechanical, organizational, technical, and ethical failures by Uber, but the scenario of striking a pedestrian crossing a street with a bicycle at night inspired dozens of entrepreneurs and engineers working on tackling such self-driving edge cases. Two such suppliers – Important and Adasky – were on the Autotech Council call and both predictably posited there technical solutions as capable of preventing the fatal crash that occurred nearly two years ago.

Important offers a cellular-based location technology capable of alerting cars to the presence of pedestrians in the path of the vehicle. Adasky offers thermal imaging uniquely suited to identifying pedestrians and other objects in the roadway at any time of day or night.

These are but two solution providers that have stepped forward to solve the Tempe-crash challenge. In the immediate aftermath of the fatal crash some analysts touted the efficacy of dedicated short range communications – a Wi-Fi-based wireless technology – as a potential means to have prevented the crash. Similarly, LiDAR (light detection and ranging) advocates leapt at the chance to assert that only their technology could solve the problem.

The fatal crash simultaneously demonstrated how close mankind had come to self-driving and how far the industry had yet to go. If something as simple and obvious as a pedestrian pushing a bicycle in the roadway could be overlooked, what else was being missed.

What was really missing was the driver monitoring piece. In the industry’s rush to remove the driver from the activity of driving, Uber engineers had overlooked, ignored, or dismissed the power and importance of the human driver. 

Humans are really good at driving – no matter what regulators may say about 90% of crashes deriving from the failures of humans. The reality is that cars ought not to crash ever. The fact that cars do crash is and ought to be understood as a product defect. Cars ought to be designed to assist drivers, who are attentive at all times but still need help on occasion – to avoid other drivers or to overcome their own vehicle or medical failures. Driver distraction can and should be detected and prevented.

Henceforth, every self-driving car presentation will have “the Uber ATG slide” showing how the presenting company would have tackled the Tempe crash scenario. The real lesson, though, was an organizational and strategic one. It was a process failure.

The car in the Uber Tempe fatal crash was not prepared properly and the driver was not trained properly. The intended use case scenario was unclear. The expected performance was ill-defined. The actual outcomes - dismal as they were prior to the crash - were disappointing and embarrassing but largely ignored.

The Uber crash also exploded the notion of a single path to autonomous driving. Decades into the process of achieving the moonshot goal of fully self-driving cars, the industry cannot agree on the ideal combination of sensors and software to meet the challenge of displacing the meat behind the wheel.

Uber ATG’s gift and legacy to the industry is an abrupt reminder of how far we have come and how far we have to go. There is no simple solution, no magical combination of sensors and software that is known to deliver full self-driving capability.  Experts tell us we must drive billions of miles to fully meet the challenge of modeling self-driving. We have but a few tens of millions of miles of self-driving under our seatbelts.

Even worse, we are still debating use case scenarios and business models and slowly grasping the need for remote control. For this, we can thank Uber ATG. 

Above all, the fatal crash was a reminder of the reputational cost of even a single failure and an indication of a low tolerance for that failure. This was an important lesson indeed in the context of 1.2M annual highway fatalities globally. We are and we should hold self-driving technology to a different standard. Human drivers are damn good at what they do, but we need help. We are no longer looking to Uber for that help - that is the price and the lesson learned from a single catastrophic failure.

Previous Post: Tesla: The Eyes Have It | Next Post: New Dawn for U.S. DOT under Biden

Let's talk

Now you know a little about us, get in touch and tell us what your business problem is.
Inquiry / Message:

please enter captcha from left