GM/Cruise leaks show them way, way behind Waymo. It's time for better metrics from everybody
GM's "Cruise" robocar unit is often cited as #2 behind Waymo. Some recent leaks of their internal metrics for progress paint a dim picture; that they aren't nearly as far along as they hoped, which does not bode well for the planned 2019 launch. In fact, they show as an order of magnitude behind where Google/Waymo was back in 2015.
The numbers that get published due to legal requirements tell us almost nothing. The methodology used by Waymo and cruse, the distance between simulated contacts, is a good start, and more teams should use it. And we should also dream up better metrics. (The simulated contact metric means every time the human safety drivers have to intervene, you build the situation in simulator, and you figure out if the car would have hit anything or not.)
In the new article linked below, I outline what this means for Cruise and what better metrics we might use. Measuring the safety of Robocars is now the biggest problem they face. We aren't that good at it yet, and even once you've done it, you have to prove that you've done it.