AutoBART is based on the assumption that autos can safely navigate the route quickly, drive close together and pull out of the right of way to pick up passengers. Here are a couple ways to test these assumptions.
There are many 2d game library's that could be used to test the control models of autos. This would be helpful to learn the dynamics of platooning and the tolerances of staying within lanes.
Researchers are already using 3d game engines like Unity to simulate environments for self driving car software. To simulate driving in BART, CivilMaps could make a 3D map by mounting a LIDAR sensor to the front of an existing BART car. Comma.ai released an app (chffr) to help learn from driving conditions. Maybe this could be used to show how hard/easy it would be to drive in BART (instead of on the road).
1/10th scale cars on a setup track
The self driving team at UC Berkeley has made an opensource self driving car (BARC) out of a 1/10th size remote control car. The parts cost $500 dollars looks like an involved setup. These could be tested or raced on a track in a warehouse.
Here is a video of one of the BARC projects.
What you'll find.
This blog is a raw dump of information about how to create the worlds best mass transit system in the Bay Area.