Researchers Instructing Autonomous Vehicles on How Pedestrians Move

author: Vineeth Joel Patel   

Pedestrians continue to perplex autonomous vehicles and the companies that are developing them. Unlike machines, pedestrians are unpredictable. Living in a city, I've seen pedestrians cross the street willy nilly and walk on the edge of the road when there's a perfectly good sidewalk close by. 

While it's easy for human drivers to notice these things and react to it, it's harder for autonomous cars. Which is why researchers at the University of Michigan are developing a system that will teach self-driving cars to predict what pedestrians are going to do next. 

Researchers at the university, thanks to a grant that was sponsored by the Ford Motor Company, were able to look into ways to teach autonomous vehicles how to "recognize and predict pedestrian movements," improving on what's possible at the moment with current technology. 

How Does It Work?

To gather data, the researchers utilized cameras, LiDAR, and GPS systems that were fitted onto driverless cars. The systems allowed the researchers to capture video footage of humans in motion. Specifically, the systems are focusing on the pace of a pedestrian's gate, symmetry of limbs, and how stable the pedestrians are when walking. 

With the snippets in hand, the researchers then recreated them in a computer simulation in 3D, allowing them to create a "biomechanically inspired recurrent neural network" to create a catalog of human movement. 

With the hard work out of the way, researchers can use the catalog to predict what "pose" and "future location" pedestrians can take when they're roughly 50 yards away from the vehicle. 

"Prior work in this area has typically only looked at still images. It wasn't really concerned with how people move in three dimensions," said Ram Vasudevan, assistant professor of mechanical engineering at the University of Michigan.

What's Possible In The Future

Why go with videos instead of still images? Apparently, a computer won't be able to recognize stop signs in the real world unless it looks at several million photos of them. Getting autonomous vehicles to spot a pedestrian and predict what it will do next is more complicated than simply identifying a stop sign. Video clips that are several seconds long will allow the software program to soak in the data much faster. 

The videos are working, as the university claims that the system can watch the first half of the short clip and make a prediction on what the pedestrian is going to do and then see if it got it right during the second half. 

The next step in the equation is getting the system to be able to predict a few steps ahead. Endowing autonomous vehicles with the ability to predict what pedestrians will do next will make obviously make the streets safer. 

Other companies have different ideas to dealing with pedestrians. Ford was testing a "visual language" system to directly communicate with pedestrians, while Lyft was granted a patent to introduce technology that would allow the vehicle to display messages to those outside of the car. 

The rush to get some sort of technology that allows autonomous cars to communicate with pedestrians in place comes after an autonomous Uber struck and killed a pedestrian in Tempe, Arizona last March. 

Vineeth Joel Patel

Joel Patel has been covering all aspects of the automotive industry for four years as an editor and freelance writer for various websites. When it comes to cars, he enjoys covering the merger between technology and cars. In his spare time, Joel likes to watch baseball, work on his car, and try new foods