AI gets lost
AI goes walk about
AI ...
You get the picture. It seems that the AI gets a lot of flak for doing what it is told to do. I was about to add to the litany when I had a second look at my route.
An AI controlled train entered a station, unloaded and loaded passengers and then instead of changing direction to return to where it came from, it continued through the station and then after some shunting to get across three tracks, it then started it's return journey.
After a closer second look, I found a direction marker left over from when that part of the track was supposed to be one way. With that removed, the AI now didn't want to traverse a set of double slip turnouts that were successfully negotiated on the way in. I added an extra signal and now the route has been running without a hitch for almost four hours with 6 to 10 AI trains interacting.
I find that within the limits of its program, the AI works pretty well. As the route designer, it is up to me to see to it that the AI has enough information to complete its tasks. That means proper signals and a path that can be followed must be provided.
The AI will do exactly what it is told to do. Now if what you said is not what you meant, that's not AI's problem. :wave:
One thing that would be nice to have is a method to make the AI wait for a signal to change or a blocking train to get out of the way. Now it tries to calculate a new path as soon as it is blocked when a little patience would solve the problem. Of course at other times it needs to recalculate on the fly. The dynamic nature of a route with multiple AI trains running means that sometimes a secondary path will work just a well as the primary. No need for a delay.
Another nice thing would be a way to rank each train (automatically or manually) so that when two trains are contending for the same switch or stretch of track, the higher rank always wins.
AI goes walk about
AI ...
You get the picture. It seems that the AI gets a lot of flak for doing what it is told to do. I was about to add to the litany when I had a second look at my route.

An AI controlled train entered a station, unloaded and loaded passengers and then instead of changing direction to return to where it came from, it continued through the station and then after some shunting to get across three tracks, it then started it's return journey.

After a closer second look, I found a direction marker left over from when that part of the track was supposed to be one way. With that removed, the AI now didn't want to traverse a set of double slip turnouts that were successfully negotiated on the way in. I added an extra signal and now the route has been running without a hitch for almost four hours with 6 to 10 AI trains interacting.

I find that within the limits of its program, the AI works pretty well. As the route designer, it is up to me to see to it that the AI has enough information to complete its tasks. That means proper signals and a path that can be followed must be provided.

The AI will do exactly what it is told to do. Now if what you said is not what you meant, that's not AI's problem. :wave:
One thing that would be nice to have is a method to make the AI wait for a signal to change or a blocking train to get out of the way. Now it tries to calculate a new path as soon as it is blocked when a little patience would solve the problem. Of course at other times it needs to recalculate on the fly. The dynamic nature of a route with multiple AI trains running means that sometimes a secondary path will work just a well as the primary. No need for a delay.
Another nice thing would be a way to rank each train (automatically or manually) so that when two trains are contending for the same switch or stretch of track, the higher rank always wins.