In one of the least surprising developments over the past few months, another road study said Michigan roads are bad, but this study is different. Not because it rated Michigan as dead last among all 50 states, but because it used millions of images from drivers across the country to determine the ratings of each road.
The study was compiled by lvl5 Inc., a San Francisco-based computer mapping company. The firm used the app Payver to pay drivers to record dashcam video from their cellphones. The company then takes those images of millions of miles of roads and uses computer vision to detect faded lane lines, potholes, cracks, and other deficiencies.
“We have almost every road in the country covered,” lvl5’s CEO Andrew Kouri said. “The point of this is to objectively measure roads."
Kouri said company staff invented the app and compiled the study for municipalities, counties and states to develop better, more frequently used grading systems for road conditions.
However, the developers also had self-driving cars in mind.
“That's kind of how we stumbled across the problem," Kouri said. "We noticed that self-driving cars are highly dependent on seeing the lane lines on the road, so if those lane lines are faded or worn down, self-driving cars might have problems where it might cross over lanes, might follow cracks in the road, instead of where it's intended to go, in between the lane lines. That's definitely a very significant problem, and having clean roads that are very easy for both humans and machines to understand is very important.”
The study's findings could have a major impact on the automated vehicle industry, said Osama Abudayyeh, an engineering professor at Western Michigan University. Abudayyeh said Michigan’s poor roads might actually be a benefit to the development of automated vehicles.
“If you put it in the harshest environments, that may make it learn quicker, instead of testing it in nice perfect conditions and then take it on rough roads,” Abudayyeh said.
Abudayyeh earned his doctorate in civil engineering at North Carolina State University, and has chaired Western's and chairman of the civil and construction engineering department for the past seven years. He said Michigan is on the cutting edge of automated vehicle technology.
“The Michigan DOT will be leading in terms of standards and other things for international implementation,” Abudayyeh said. “We are the biggest manufacturer of automobiles. We tend to be the state that people look at as the lead, and we are the lead.”
Abudayyeh points to recent events as setbacks for self-driving cars, but said the technology will take time to develop.
“What I have seen, the estimates are between 2020 and 2035 when more and more of these vehicles will become commonplace. It will take time to perfect the technology,” Abudayyeh said.
Faded lane lines and potholes present the same problem, Abudayyeh said.
“These vehicles are based on sensors, and sensors are cameras that look for markers on the roads, and those markers get hidden under the snow, there needs to be more development in the way you do image processing from the data you collect from these sensors,” Abudayyeh said. “The more challenges there are, the better for these vehicles to drive in and test. These vehicles need to learn from the environment in which they test. And I'm hoping that's what happens, as the car is on the road more and more, it learns on its own and builds enough database for different situations.”
Abudayyeh is OK with the pace that the industry is developing, and said it will take the toughest safety standards and possibly a new generation of people who are more in touch with automation to make self-driving cars commonplace.
“I think technology will be coming, it's just slow for obvious reasons," he said. "Safety is very important. No need to rush and have more accidents, and we should take it step by step."