Making AVs Safer
In an effort to make AVs safer, researchers at the Driving Safety Research Institute at the University of Iowa wanted to learn the challenges they encounter when operating on rural roads. Watch this video to see what they learned after more than three years of testing.
“We’ve chosen to own our ruralness,” says Daniel McGehee , director of the DSRI, which was the first in the nation to study automated driving in rural conditions. “We want to look at increasing people’s quality of life through adding another dimension of transportation that’s not available today, that will allow rural folks to extend their lives in a social and healthy way.”
Transcript
00:00:00 - [Omar] This project looks at the challenges and opportunities of operating an automated transit vehicle on rural roadways. - (Daniel) More traditional automated driving systems rely on computer vision to look at curbs and paint in the roadway to position the vehicle in the middle of the lane. We are the first place in the world to do automated driving on rural roads that are unmarked and on rock roads or gravel roads.
00:00:28 So that's something that's very unique to the University of Iowa. Traditional GPS systems that are on our phones and in our cars, are generally accurate to plus or minus about one meter. We use a next generation GPS called high definition GPS, that gets us down to plus or minus a few inches. We can triangulate a number of sensor data to be able to position the vehicle down the center of the particular lane that we wanna be in.
00:00:57 - (Cheryl) The route that we chose was really designed to challenge the vehicle. We wanted to understand where the limitations of the automation might be. One of the things we learned pretty quickly on was that the gravel road was mapped to be two lanes. When we were driving on the gravel road, we drove almost in that loose gravel where nobody drives. Because in rural Iowa, we drive in the middle and then when a car comes, we kind of veer off.
00:01:21 So that was something that we identified pretty early that they would need to change how the vehicle behaved, whether we'd change the map, or we'd change the behavior of the vehicle. - An automated vehicle tends to be very rule-bound. How do you go about teaching an automated vehicle to sometimes break the rules, but also know when it's not safe to do so? - While a smaller segment of the population lives in rural areas,
00:01:46 nearly half of all traffic fatalities happen in rural areas. So if automated vehicles can realize their promise of improving safety- - (Computer-generated voice) Slower vehicle ahead. - ... we can make an incredible impact in terms of reducing the loss of life and reducing injuries for everybody. - In the future, we'll be looking at machine learning techniques, so that automated vehicle will be continually learning
00:02:12 about all these different unique scenarios. And that's one thing that we do here is concentrate sometimes on things that are very hard to do to understand what the next generation of automated driving is all about.