Strict AV Evaluation is Needed to Promote Consumer Acceptance

The future of mobility appeared flawless right up until the dream shattered with a fatal Uber accident caused by both human error and machine malfunction.

It was an incident that sent shivers up the spines of everyone inside the auto industry, made even worse when it was followed by a fatal accident relating to Tesla’s Autopilot function. Human error, perhaps even neglect, was a factor in that accident as well but it was no less troubling for those who want to bring autonomous vehicles to market.

Huei Peng, director of Mcity, an AV test track at the University of Michigan, is hoping to help prevent similar accidents from occurring. He and other researchers have devised a concept for evaluating the safety of highly automated vehicles. The goal is to assess vehicle safety in advance and encourage automakers to voluntarily evaluate their automobiles before road tests begin. “After the Uber and Tesla crashes, there has been a lot of concern from the general public,” said Peng. “The question is: why is the government allowing these vehicles on public roads?”

Peng isn’t looking for governments to quash innovation and halt AV testing. However, he thinks it’s time for minimum performance checks, whether from the school or another entity, to independently verify a certain set of standards.

“Right now there’s no constraint, no regulation,” said Peng, adding that the industry is worried about the public outcry and a possible backlash that could follow future accidents. He sat down with Mcity’s Leadership Circle companies, which includes Toyota, Ford, General Motors, Intel and Aptiv, among others, to explore what could be done.

Time change

This led to the creation of a new concept, the ABC test (Accelerated evaluation, Behavior competence and Corner cases) for independent safety assessments. The process begins with “important sampling”, an idea that ditches the notion that millions of miles are needed to reach deployment. In this instance, only the “interesting” miles would be sampled, rather than the “boring, empty miles” that occur most of the time.

Next comes the 50 “basic behavior competence” scenarios that were derived specifically for this test. The scenarios include things like high-speed lane merging but not all vehicles have to pass all 50 scenarios. For example, shuttles from Navya and May Mobility do not exceed 25mph and would therefore be allowed to skip any high-speed tests.

Lastly, Peng looks at the corner cases to better understand any anomalies of the cars in question. “Everybody, including Uber and many other companies, learn lessons the hard way,” he said. “Autonomous emergency braking, designed to mitigate situations exactly like this crash, was turned off. If it had not been turned off, it likely would have been triggered and likely the impact speed would have been reduced.”

Just as there are challenges in building autonomous vehicles, there are difficulties in testing them as well. Without any governmental authority, tests will remain voluntary for the foreseeable future. Some organizations have chosen to purchase products to test their safety but there aren’t any driverless cars available outside of the low-speed shuttles from Navya and EasyMile. This can make it very difficult to get a consensus on AV safety.

Too many fatalities

Peng is very aware of the media outrage that has followed the fatal Uber accident but he takes issue with the lack of attention toward the other deadly accidents that occurred on the very same day.

He said: “On average about 100 people get killed every day but only in this case do we say, ‘You shouldn’t be watching videos.’ How about the other 99? Many of them are human error. ‘You were texting, weren’t paying attention, daydreaming,’ whatever. Road rage – people do that all the time. Hopefully with another pair of eyes, the robot, whether it’s a copilot or chauffeur, is in theory improving continuously. We know humans are not.”

One day, autonomous cars might even be able to outsmart human drivers. “People would say, ‘Robots will never be intelligent enough to be the chess champion or Go champion,’” Peng added. “Robots have now done both. How about Jeopardy? The robot understood the verbal question and found the answer before the human? If you think about it, that’s pretty impressive. Personally, I think many of us are in agreement that robots are not better than human drivers today but we all agree the potential is there. We just don’t know how long it will take.”

Cars are a bit more complex than a TV game show, however. They move at high speeds across a wide range of environments, all the while interacting with other automobiles, pedestrians and cyclists. “Most of the time it’s hassle-free,” said Peng. “In an empty road, everybody behaves. That’s not a challenge.”

When everyday driving conditions are put into place, it introduces all the problems that are present on public roads today. These are issues Peng hopes to help automakers solve in the years to come.


Leave a comment

Your email address will not be published. Required fields are marked *