Waymo Spins Tragedy Into Triumph for Autonomous Vehicles

Last month, a motorcyclist was injured by one of Waymo’s self-driving Chrysler Pacificas. According to the accident report, a car in the left lane attempted to merge into the same middle lane as the Pacifica test platform, which was operating in autonomous mode. The safety driver then “took manual control of the Pacifica out of an abundance of caution, disengaged from self-driving mode, and began changing lanes into .”

Considering the AV wasn’t traveling above 25 mph, it’s a little curious the driver took evasive action, unless the second car attempted to merge directly into it. Regardless, the Pacifica’s lane change placed it into direct contact with a motorcycle that was moving slightly faster. Waymo said that, had the autonomous system been left in play, the vehicle would have assuredly avoided the accident.

Of course it would have.

Waymo conducts software simulations after each instance where a driver takes control of a vehicle to see how the software would have responded.

“Our review of this incident confirmed that our technology would have avoided the collision by taking a safer course of action,” explained the company’s CEO, John Krafcik. “While our test driver’s focus was on the car ahead, our self-driving system was simultaneously tracking the position, direction and speed of every object around it. Crucially, our technology correctly anticipated and predicted the future behavior of both the merging vehicle and the motorcyclist. Our simulation shows the self-driving system would have responded to the passenger car by reducing our vehicle’s speed, and nudging slightly in our own lane, avoiding a collision.”

It’s still a shame the motorcyclist had to eat the bumper of a vehicle owned by the tech firm and end up in the hospital, but, boy oh boy is it ever a big relief to know it wouldn’t have happened if a human wasn’t in control of the car.

Unfortunately, the reality is that Waymo’s platform isn’t omnipotent. There was another incident in May that the company’s software couldn’t have accounted for. That incident, which took place in Chandler, Arizona, is a little complicated, however. Initial police reports claimed the test platform (another Pacifica) was operating in autonomous mode when a Honda Civic swerved into its lane to avoid another vehicle. Judging from the damage and report, it appears that the Honda turned directly into the side of the minivan.

There’s more anecdotal evidence that these platforms aren’t yet up to snuff. In August, residents of Chandler, Arizona told The Information that they hated having to share the road with Waymo’s test vehicles. Numerous locals have complained about the vehicles making erratic stops and struggling to get through turns involving multiple lanes of oncoming traffic, even after being stopped at a light.

For what it’s worth, most of this is a byproduct of Waymo programming the cars to act as defensively as possible. AVs may have come a long way, but they can still get confounding by new information; humans are much less attentive but more skilled at adapting. This causes autonomous cars do to things that seem stupid for no apparent reason, whereas a human driver wouldn’t have become perplexed in the first place. We tend to ignore useless (and sometimes important) information, whereas a computer still has to process everything and decide what it means.

Does that mean the Pacifica would have avoided the motorcycle on October 19th if left to its own devices? Maybe. But we also know that tech firms and automakers are hesitant to admit to gaps in their hardware. Make no mistake, as incredible as the technology is, it’s perpetually overblown by those developing it. They’ll always tell you it works unless faced with a mountain of evidence to the contrary.

[Image: Waymo]