Building Trust in Automated Vehicles Will Take Many Tools

In September 2015, news reports emerged of a wide-ranging problem with the emissions of diesel engines from vehicles built by the Volkswagen Group. Millions of vehicles globally were emitting significantly more nitrogen oxides than had been indicated during certification tests. This issue was enabled by some hidden features in the software. One of the lessons we learned is that it is remarkably easy to hide cheating in computer software. As vehicles become more software-defined, especially with the onset of automated driving systems (ADS), we must take tremendous care in how we validate systems where the consequences of failure could be fatal. 

As someone that came out of school with a mechanical engineering degree, if I’ve learned anything over the past 30 years, it’s that software can be used to enable all sorts of functionality that cannot be achieved through purely mechanical means. ADS is absolutely one of those capabilities. Companies like General Motors dreamed of ADS back in the 1950s, but the technology simply didn’t exist yet to make it a reality. In 2021, we now have vehicles on the streets of multiple cities around the world carrying passengers and delivering goods without human drivers. 

Numerous surveys have shown that a majority of people are not comfortable with the idea of ADS, and many say they will never trust it. That’s a perfectly reasonable response given that most people have never seen an automated vehicle (AV), much less ridden in one. Those attitudes often shift dramatically after a real-world experience. However, a lot of work still needs to be done to build trust in this technology. But given the relative ease that has been demonstrated in cheating on standardized tests through software, that won’t be easy. 

Present the Evidence

All of the companies developing ADS are simulating tens of billions of miles of driving with their systems in a variety of scenarios in addition to the millions of miles of on-road testing being accumulated, including robotaxi pilot programs. Waymo researchers have just published a white paper on some new simulation work in an attempt to demonstrate that its ADS is safer than human drivers. The team collected data from 72 fatal crashes in the Chandler, Arizona area where its Waymo One robotaxi service operates. Working with accident reconstruction experts, they simulated the same scenarios replacing the vehicles that caused the crashes (called initiators) with Waymo vehicles. Where multiple vehicles were involved, they also ran the same tests with the responder vehicles that were struck. 

The results showed an advantage for the Waymo Driver ADS. When the ADS replaced the initiator, 100% of the crashes were avoided. This was attributed to a variety of causes, including obeying speed limits and signals and not driving drunk or distracted. When the ADS was substituted for the responder vehicles, all of the crashes were avoided or the impact was reduced significantly. In addition to following the road rules, the ADS has better situational awareness than a human. 

By no means is this proof that AVs are yet entirely safer than human drivers. These scenarios were in a single city with good weather and more visible intersections than a dense urban environment like San Francisco or Miami or in inclement weather like Detroit in January. However, this sort of effort is key to both validating the efficacy of ADS and garnering trust from the people who will coexist with the technology.