The basic idea of a successful AI in railway simulation is fairly simple: Do as the prototype does. Train movements follow rules. Dispatching follows rules. Apply these rules, combine the two, and your AI will be working. That's what Zusi did from the beginning. I admit you need a thorough understanding of the prototype for this approach to succeed. From their background, mainstream game developers may have a different view on AI, and I guess that's were the problem starts.
I think Trainz is much closer to Zusi here than to Kuju/MSTS or Kuju/RS-RW. in Trainz, there is no fundamental difference between player and AI trains, signals work properly, all trains observe signal aspects, track markers fill in gaps.
What Vern is missing in Trainz is the "working timetable" (British lingo). In Zusi this is a component which combines infrastructure and train data, very much like the prototype again. In my impression Trainz scenario building leaves the prototype path, in favour of something that may produce more play value. The question is: Play value for whom? I am aware that the Trainz approach with individual shunting or a working industrial production chain provides excellent entertainment to many users. Not very prototypical but who cares. Nevertheless, the foundation for creating a working timetable does exist in Trainz, it's only not exploited at the moment, as I see it.
