In an effort to demonstrate its high levels of autonomous driving, Waymo likened dozens of real-world fatalities that took place in Arizona for nearly a decade. Google’s spinoff company has discovered that replacing any vehicle in a two-vehicle car crash with robot-driven minivans could nearly zero out all deaths, according to data released today.
The results are aimed at reinforcing Waymo’s hypothesis that autonomous vehicles operate more safely than those driven by humans. With millions of people dying in car accidents worldwide each year, autonomous vehicle (AV) companies are increasingly relying on this safety case to push regulators to pass legislation allowing more fully autonomous vehicles on the road.
However, this case was difficult to prove, thanks to the very limited number of autonomous vehicles currently operating on public roads. To provide more statistical support for what it supports, Waymo turned to counterfacts or “what if?” scenarios, aimed at showing how robotic vehicles would react to real situations.
Last year, the company released data on 6.1 million driving miles in 2019 and 2020, including 18 collisions and nearly 29 cases of collisions. In those incidents where safety operators took control of the vehicle to avoid a collision, Waymo engineers liken what would happen if the driver did not disable the vehicle’s self-driving system to create a counter-example. The company has also made some of its data available to academic researchers.
This counter-example work continues in this latest data release. Through a third party, Waymo collected information about every fatal accident that occurred in Chandler, Arizona, a suburban community outside Phoenix, between 2008 and 2017. Focusing only on the conflicts that occurred in its field of operational planning, or in the area of about 100 square miles in which the company allows its cars to be driven, Waymo detected 72 collisions for simulation representation to determine how its autonomous system could respond to similar situations.
Some of these accidents involved one vehicle, while most involved two. In collisions with two vehicles, Waymo conducted separate experiments simulating its autonomous vehicles in the role of each vehicle – first replacing the vehicle that initiated the collision and then replacing the vehicle that responded to the actions of the other vehicle. In collisions with a vehicle, Waymo likened the unique vehicle. This led to 91 simulations in total.
- VW to invest in Greece for autonomous driving tests
- Tesla FSD Beta reaction with cat crossing the road
The company re-enacted these accidents, systematically aligning the vehicle’s trajectory to make sure waymo vehicles are exposed to a similar situation than in the actual fatal collision. Waymo used the same simulation platform it uses to train and evaluate its autonomous vehicles on virtual roads for routine operations.
The results suggest that Waymo’s autonomous vehicles would have “avoided or mitigated” 88 of the 91 total simulations, said Trent Victor, director of safety research and best practices at Waymo. In addition, for mitigated accidents, Waymo’s vehicles would have reduced the likelihood of serious injury by 1.3 to 15 times, Victor said.