TreeSharp and data driven behavior trees

[GliderPro]

I've been experimenting with TreeSharp lately and reading up on behavior trees. There are several articles recently which talk about these new data driven behavior trees. Is anyone here working with them? Are there any new AI methods out there that you are working with in your projects?

[streppel]

I'm using apocs behavior tree implementation in my bot. The honorbuddy wiki is a nice resource for getting started in it. If you have any questions, just ask and I'll try to answer, and the others will do so for sure.

[Bananenbrot]

IMHO, a behavior tree for single agent AI is overkill. See Game AI Forums for my details. 'data-driven' seems worthy, but turnaround times are extremely short when working with a proper AppDomain loader/unloader in C#, which defeats the 'not scriptable' part of choosing a BT over regular driven-out interfaces.

[GliderPro]

Bananenbrot, Can you explain a little bit about what you mean by coroutines? All of my bot logic has been done with state machines so far. I'm looking for a method which makes bot logic easier to maintain/edit.

[Bananenbrot]

By coroutines, I just mean a way to stop execution and continue at a later point (see Coroutine - Wikipedia, the free encyclopedia). That's essentially the RunStatus.Running return value, which should be realized (if ever used) with a custom enumerator, e.g. a method which uses yield return RunStatus.Running;. You almost never need it, because in a single agent AI there will never be a moment where you need that extra tiny bit of performance and it eliminates the benefit of automatic selection of the right child node (because that specific node does not know if there are any nodes with a igher priority).
Most people have their logic in 2 or 3 big files, which make direct references to custom classes etc. They forget about pulling out coherent interfaces, so they can switch out behavior at runtime if needed. Combined with unit test, I can pretty much verify every tiny bit of behavior on its own and then make some (unfortunately manual) regression test while looking at how smoothly the bot runs.