I think we can all be sure that once this is perfected, in real life there will be far fewer collisions per year with trains than the way the model performs rig

I think we can all be sure that once this is perfected, in real life there will be far fewer collisions per year with trains than the way the model performs right now. When people are stupid enough to put trains at street level (on roads previouly optimized for cars) you can't later expect AI to "fix stupid" on its own when the best traffic engineers have never even achieved very ideal avoidance of train collisions like this compared to elevated or subterranean track foundations.