News and culture through the lens of Southern California.
Hosted by A Martínez
Airs Weekdays 9 to 10 a.m.

How self-driving cars detect pedestrians




Uber has a self-driving car partnership with Volvo.
Uber has a self-driving car partnership with Volvo.
Uber

Most Americans are already suspicious of self-driving cars. Now those safety concerns are on the front burner again after an autonomous car hit and killed a pedestrian in Arizona over the weekend.

How to detect and avoid pedestrians and bicyclists is one of the biggest sticking points for self-driving technology. Anuj Pradhan is a researcher who specializes in pedestrian detection and autonomous vehicles for the University of Michigan Transportation Research Institute. He joined Take Two's A Martinez to explain how these detection systems work.

Why it's so difficult for self-driving cars to detect pedestrians

In the broader general sense, detecting pedestrians and other vulnerable road users by automated vehicles is a challenging task. There’s a lot of R&D going on in terms of detection. I would say if somebody has the full suite of sensors -- radar, lidar, cameras -- detecting the presence of pedestrians, peole have gotten good at it. What they’re not good at yet is inferring the intent of a pedestrian once it is detected.

How self-driving cars perceive pedestrians

When we as human drivers look at pedestrians, we use eye contact, which is a very, very major piece of communication between the pedestrian and a driver. A driver looks at a pedestrian’s gait, speed, posture and a whole host of other factors, intuiting how old or young they may be, if they’re in a group. All of those are processed by a human driver to figure out the intent of the pedestrian. These have to be accounted for by automated systems, and it isn’t being done yet.

What a self-driving car sees that says it’s a person

It depends on what sensors the autonomous vehicle has. If an AV has a suite of sensors which include radar and multiple cameras and lidar, then it will give the automated system a very good idea of what’s out there. A combination of sensors makes it possible to have redundant information because if your cameras are not working well because it’s night, radar or lidar will pick it up. The sensors give a good sense of what’s around the car — of what exists in terms of other vehicles, pedestrians, vegetation and so on. They’re pretty good at it.

Pedestrian detection isn't just a technology problem

The road is shared, so regulastors will have to get involved. We will have to hear from urban planning folks and folks that are not necessarily involved in the technology aspect of self-driving cars. This is not just a technology issue, it’s a societal issue.

Should we be freaked out about this self-driving car fatality

All of these deployments are test deployments, so it’s all a big experiment. What should be paramount to everybody is safety. In terms of people who have to interact with these systems that are out there on the streets who don’t want to but have to, what one would say is, 'Well, you are participating in a large, large experiment.' We can hope that everybody who's involved in conducting these take into account all the safety aspects, but it’s still an experiment.