This is very unfortunate for Pittsburgh and Aurora. People will be injured and killed for absolutely no reason. And yes there is a far better way.
It is a myth that public shadow driving is a viable method to create AVs. That process can never come close to creating an AV. You cannot drive the one trillion miles or spend over $300B to do so. You also cannot run thousands of accident cases thousands of times each or have any more casualties especially a child or family. As Uber, Aurora, Argo etc drive more complex scenarios and run thousands of accident scenarios many, many people in Pittsburgh will be harmed or killed for no reason. It is unavoidable on this reckless and fruitless path.
Someone explain to me why simulation cannot be used to replace 99% of public shadow driving for development and testing? Yes real-world data (with hands on the wheel) is critical especially to inform and validate the simulation. But exactly what can I do in the real-world I cannot in proper simulation? On the flip side exactly how am I going to learn/test complex and accident scenarios in the real-world, especially those involving accidents and variables that are hard to produce once let alone hundreds or thousands of times?
Impediments to Creating an Autonomous Vehicle https://medium.com/predict/impediments-to-creating-an-autonomous-vehicle-d3cfee299749
The most disappointing and hopeful part is Urmson’s evolution. He understands and has had the courage to say metrics like miles driven are misleading. And that handover is dangerous and should not be used. Yet he uses public shadow driving for development. My assumption is he doesn't have experience with aerospace/DoD/FAA level simulation or engineering practices. While they won’t assure full L4 they make it possible. It is not remotely possible to do using public shadow driving.
Chris Urmson — Please Step all the Way Up