We still live in the early days of autonomous vehicles. We’re toddlers, technologically speaking, and our algorithms are still learning to walk, er, drive.
That’s the impression you get from reading the various recaps of an autonomous car test run that General Motors hosted for journalists this week in San Francisco. Wired’s Aarian Marshall noted that the company’s self-driving Chevrolet Bolt, using software from GM’s Cruise Automation division, gave a “herky-jerky ride” despite having made progress since GM acquired the San Francisco-based startup for big money in 2016.
Similarly, Recode’s Meghann Farnsworth wrote that the autonomous vehicle came to a complete stop in the middle of the road after a dog on a leash paused in the middle of a crosswalk to relieve itself. The car waited for the dog to finish its business, rather than passing around it, Farnsworth reported, consequently giving its passengers an unexpected show.
Certainly, anecdotes like this are tantalizing to note by any observer of new technology—particularly technology that would replace something we humans do (and often enjoy doing). It must be satisfying for the public relations teams at companies like GM, who are seeking to stir conversation—almost any conversation—about the wild possibility that their thinking vehicles might transport us in the not-too-distant future.
Of course, research into autonomous vehicles is continuing on even in places with far smaller PR teams than GM, Tesla, Google’s Waymo, Uber, or Lyft. These efforts tend to target a niche sector, sometimes with a different technical approach and rollout strategy. In San Antonio, TX, a group at the Southwest Research Institute (SwRI) is developing autonomous vehicles based on algorithms that process images of the road below the vehicle almost instantaneously; the car follows a set route based on nuances in the road, almost like matching a fingerprint. The institute also uses lidar and radar systems to observe what objects are near it, like most self-driving systems.
The research institute’s system has its limitations. A driver must take the vehicle on any route they hope to have the car one day ride on autonomously so that a camera below the car can record the road, creating the map to follow later. That means a lot of driving, and a lot of data to process. (It also means problems if the gets repaved, though researchers at the institute say the tech will still work even if about 70 percent of the car-width image is blocked.) But the researchers contend that their system allows for a much more accurate route—with only one to two centimeters of variation—allowing passengers to feel comfortable that the car isn’t going to veer out of a dimly marked or unmarked lane.
The reality is that a system like the one being developed at SwRI might be more useful for the military or the trucking industry than consumers.
“There’s some interest in the industry to [use] this for automated convoy,” Mark Alban, a research analyst at SwRI who was one of the developers of the technology, said during a tour of SwRI’s autonomous vehicle facilities earlier this year. “One of the main objectives is to improve fuel economy. A big issue is having the vehicles be right behind each other… It makes a big difference if they’re really well aligned.”
Despite the intricate details of the technology, the part of the tour that sticks out in my mind—and the part that everyone always asks about—was simply comparing how a computer drives to how a human drives. SwRI’s autonomous vehicles—one sedan, and one semi-truck—largely gave smooth rides, though it’s easy to note the differences from a human driver.
While a human might ease into a complete stop, the robot car can sometimes give a harder brake. And potholes. A computer, unless it is programmed to avoid a hole in the ground, will ride over the same hole time and again until it is told not to.
To an extent, the media’s tendency to focus on the driving skill of unmanned vehicles must delight the companies that are developing them. It takes our attention off the fact that there are not yet regulations around ensuring these cars can be safely deployed. And it keeps us from talking about the business models—or lack thereof, as Xconomy first pointed out in March—that may end up being a mirage for some of the field’s participants.
GM says it plans to have a self-driving line of ride-sharing cars available by 2019, which would place it ahead of Ford, which has similar goals, according to The Verge and other news reports. But questions around the economics of those business goals—the costs of operating and maintaining large fleets of cars, and potential losses from not getting the full use out of them—still linger.
