The approach is to replace the other with weaker examples being replaced by "progeny" that are copies of the better-performing networks with slightly tweaked parameters (just as a child is not a perfect clone of its parent). This is the way to get rid of the poorer-performing networks while they are away.
There is a risk that the method is focused too much on short -term improvements. To fight this, Waymo created "niches" where neural networks challenged each other in subgroups to get strong results while preserving diversity that could be better suited for real-world driving conditions.
The results were promising when applied to pedestrian detection. The PBT approach dropped false positives by 24 percent, even though it took half as much time. The experiment went so well that Waymo has been using PBT across other models. That, in turn, promises self-driving cars that can better cope with the complexities of driving and avoiding collisions.