Teaching Autonomous Cars to Drive in Parking Lots Is a Tricky Test for Waymo

Home > News > Content

【Summary】Parking lots are some of the most complex areas and are proving to be incredibly difficult for self-driving vehicles to master. To tackle the issue, Waymo is testing its autonomous cars’ ability to handle parking lots on 91 acres at a former Air Force base.

Original Vineeth Joel Patel    Aug 18, 2019 6:30 AM PT
Teaching Autonomous Cars to Drive in Parking Lots Is a Tricky Test for Waymo

Parking lots are some of the most dangerous places for pedestrians and automobile passengers alike. While parking lots are tricky for human drivers, they're even more of a handful for autonomous vehicle. Unlike roads with well-defined lines, clear road signs, and a regular flow of cars, parking lots aren't as clear cut. 

For example, pedestrians cut across roads, drivers navigate without heading signs, and traffic is much more concentrated. Waymo provided Popular Science with a look at how it's preparing its autonomous vehicles for parking lots and it boils down to practice.

Testing, Testing And More Testing
"Parking lots are uniquely challenging from surface streets," Stephanie Villegas, structured testing lead at Waymo, told the outlet. "They don't really have standardized rules for how people should and can move about within them." Anyone that's been in a Costco parking lot on a weekend can attest to just how lawless parking lots can be.
But it's not just the lawlessness that's tricky for driverless cars to understand. There are numerous social cues that humans use to communicate with one another in parking lots. If we see a spot open up, we might holler, wave, or signal to see if someone is going to vacate that spot. Autonomous cars can do this. At least, not instinctively.
To make up for the lack of social competence, Waymo is testing its driverless cars rigorously on 91 acres of road at the former Castle Air Force base near Merced, California, claims the outlet. With that much space at their disposal, Waymo can create parking lots with all sorts of obstacles for autonomous cars to be put through tests.
"Each time you turn the dial to increase complexity, the vehicle is beginning to have to assess and predict the behaviors of multiple things at the same time," said Villegas. "That is challenging for sensors."

Maps Are Another Crucial Part
Practice isn't enough, though. Popular Science claims that engineers closely monitor how the vehicles complete the tests and then make any necessary changes to their software when needed. 

Dumpsters, for instance, were difficult for autonomous vehicles to comprehend. Large trash bins are usually near areas that have swinging doors and concrete walls, which are difficult for autonomous cars to comprehend.
Practice and tests are only two parts of the equation – maps are another major part. Just as with large cities, Waymo maps out parking lots using their own autonomous vehicles' sensors, lasers, radars, and cameras. This allows the company to get all of the necessary data, which includes height of curbs, stop signs, and physical size in a 3D space. 

After getting the maps down, engineers go in and remove dynamic elements, like pedestrians, that won't be there the next time the driverless car drives through. Good maps are also important, as engineers can go in and designate areas that might be off limits, like service areas, loading docks, etc.
It may sound like a familiar process to getting an autonomous car to drive on a street, but being able to master the intricacies of getting around safely in a parking lot is even harder. 

All of this reveals just how difficult it is to get self-driving cars to behave like human-operated machines.

Prev                  Next
Writer's other posts
    Related Content