AI is worthless. I ran some tests on my CSX River Line route.
I placed two locomotives; one approaching a permissive intermediate signal, and the second, about 1/2 mile past that same signal, in the next block. All blocks on my route are now signaled.
I placed a trackmark, and set both trains to drive to the same mark.
I used the default UP SD40's.
Test 1: Using Safetran Clight 05 signals:
The trailing loco approaches and stops at the signal. It does not proceed into the occupied block. Instead, it waits until the leading locomotive is out of that block, and the signal turns to approach.
The trailing train then passes the approach signal, and instead of proceeding at 1/2 track speed, it accelerates to full track speed.
The trailing loco remains at full track speed until it gets to about 1/2 mile of the next signal, which is now approach (the leading loco is through the block and into the next).
Again, the trailing loco passes the approach signal at reduced speed, then immediately accelerates to full track speed.
Test 2: Changed signals to USA 05
Same setup as Test 1, but this time I left the leading loco alone; did not give it instructions to move, so it is sitting about 1 mile past the signal.
The trailing loco approaches the red signal, and continues past the signal at reduced speed. It then accelerates to full track speed until it gets very close to the stopped locomotive. In real life, there would have been a crash!
The trailing loco does stop before hitting the lead loco, but it decelerates more like a truck than a locomotive.
Now, I give the lead locomotive the command to drive to the same trackmark the trailing loco is driving to.
After there is some distance between the two locos, the trailing loco starts moving again. At first, it runs at reduced speed, but as soon as the gap between locos is farther, the trailing loco speeds up to full track speed.
This performance is utterly useless. It would appear to me that AI is completely broken, with respect to permissive signals; except for those signals which Justin has taken so much time and effort to code so that they will work more appropriately with this fouled up game.
Justin, I give you all the credit and none to Auran for at least making signals that are usable with AI.
Somehow, I feel that I have been here before. I recall back in TS2009 I had exactly the same problem. The only signals that work respectably with AI are those by Norfolksouthern37. We may not get the SAP operation, but at least we don't have trainz running into each other all the time!
I am afraid that my (and others) requests to fix the signaling system in Trainz has fallen on deaf ears. Auran is not very interested in a few of us who are knowledgeable in railroad operation and will even notice that signals don't work prototypically. I think we can all see that in the release of TS2010 compared to TS2009. There are very few functional improvements; most of the improvement was spent on "eye candy" and making the game more "user friendly" for gamers rather than creating a more realistic railroad simulator.
All that said, I wouldn't give up my Trainz for anything. It helps to relax me when I'm stressed (as long as I don't have freakish things happen to me in the middle of a session in Surveyor

).
I like driving my trains myself, and rarely use AI. It's a challenge handling 4 or 5 trains, but it's about the only thing I can multitask 
Once again, let me thank Justin for the work he has and continues to do on his content. I am ready, willing, and able to do any further testing you need.
FW