SACRAMENTO — Though scant on details, a report released by the DMV Wednesday shows that driverless cars have a long way to go before they can get rid of humans, experts said.
The data from the Department of Motor Vehicles includes autonomous vehicle test results from 11 companies. The information includes details of when drivers had to take control of the cars, either because the test driver felt uncomfortable or due to a glitch in the technology. The data only includes miles traveled on public roads in California, and doesn’t include testing at private facilities or outside the state.
“It’s a small glimpse into a small part of what a relatively small number of companies are doing,” said Bryant Walker Smith, a scholar with Stanford Law School’s Center for Internet and Society.
There are more questions that need answers, Smith said, such as what would happen if the driver didn’t intervene. Would the car know to pull to the side of the road to avoid a collision?
“I’d also want to know how ‘bad’ the hypothetical outcome avoided by a driver-initiated disengagement needs to be for Waymo (for example) to include such a disengagement in its count,” he said. “In other words, if the driver hadn’t intervened in any given instance, would a crash have necessarily resulted?”
Delphi Automotive Systems, for example, reported several instances where the cars could not read traffic signals due to “poor sun conditions,” or when the cars had trouble changing lanes during heavy traffic. Google’s Waymo, on the other hand, said the human drivers had to step in most often due to software discrepancies, followed by an unwanted maneuver of the car or the reckless behavior of another driver. Other times, humans took over because there was heavy pedestrian traffic or out of extra caution for a cyclist sharing the road.
The reports have “nowhere near the kind of data” needed to understand how the companies compare with each other, since driving conditions were different for each company, he said. Highway driving on sparsely populated roads, as was the case for Ford, which only tested its vehicles on Interstate 10 between Los Angeles and Arizona, is much less complicated than driving on urban streets, as was the case with Mercedes-Benz, he said.
Of the 11 companies that filed reports, two of them — Honda and Volkswagon — said they never tested the vehicles on public roads. The others ranged from 530 miles logged by Tesla during only one month in 2016 to the 635,868 miles traveled by Google’s Waymo vehicles. The reporting periods also depended on when each company secured permits to begin testing, with some companies reporting more than a year’s worth of data and others with just one or two months.
A more interesting comparison would be to see how the companies are doing over time, said Steven Shladover, the program manager for mobility at the UC Berkeley Partners for Advanced Transportation Technology (PATH), a research and development program.
For several companies, at least, the reports did show progress. Waymo reported a 75 percent drop in its rate of disengagement — defined as when a human took control of the vehicle, and Consumer Watchdog’s John Simpson said it was clear improvements have been made.
But, that doesn’t mean the technology is ready to leave humans behind, he said.
“There still is a significant number of real world situations where the vehicles can’t cope,” he said. “They are getting better, but they have a long way to go.”
