Self-driving cars can accomplish the same thing as local trains, even better. In "local mode" they get you to and from your home/job/shopping mall to the main artery, which you'd need to do with rail anyway.

Then once on the artery (freeway) they can drive a lot more efficiently than human drivers... a lot faster, closer, and without any need for traffic lights. Instead of stop and go, they would simply coordinate with each other (or be centrally coordinated) to go at a speed that maximizes throughput while allowing on/off ramps to function smoothly.

That would buy us a lot of bandwidth. Very long-term, the solution is probably under-grounding most local roads in metro areas, so the surface would be more green and park-like (also frees up real estate for housing). Not cheap or easy, but could be done in the very long haul. Electric vehicles make it more practical (no emissions to ventilate). Such a network could designed from the ground up for smooth autonomous flow. Your garage would be underground, or more likely rather than owning one, you'd just catch the next available pod, so it would work like a train but with individual routing and timing. If not stops at each house, there would be one every block or two.

The challenge to that is that AI is nowhere near ready to replace human drivers in the real world, and nobody even has any idea how to get there. The workout around to that is freeways modified to accommodate self-driving cars operating within set parameters, ie remove the unexpected factors which AI can't cope with. A virtual concrete rail line if you will. Probably require autonomous mode on freeways, at least during rush hour.

You'd probably have to drive manually on the surface streets, AI can't deal with the myriad unexpected factors you find in the local neighborhood (unless you totally ban all pedestrians, kids pets, bikes, scooters, human drivers, etc).