I've been thinking on what consequences would arise on Hierarchical task network planner if it's used with Multi-threading


HTN: http://www.gameaipro.com/GameAIPro/GameAIPro_Chapter12_Exploring_HTN_Planners_through_Example.pdf



So far my thoughts are the following


Pros:

- Planners can find the list of tasks faster using multithreading

Cons:

- When changing the world state, it needs to lock the variable so that race condition would not happen.


I've tried google around, but no luck in finding HTN planners being used in a multicore engine.
Please let me know if there's an efficient way to do this.

It doesn't sound like a good candidate for multithreading to me. Tree-style problems are not great for multithreading - they start off with fewer tasks than threads, then quickly have more tasks than threads, and yet still have strong dependencies between the tasks.

But you don't need to lock the world state more than once as you don't need to change it - each part of the plan would need a copy of the world state because it works with hypothetical states, each based on changes caused by earlier parts of the plan.

If you can get a single copy of the current world state at the start of the process, then no problem - you can just work from there in an AI thread.

You might want to consider the possible ramifications of having a planner that returns sooner on faster PCs; does that mean the game gets a little bit harder the more powerful your machine is?

Modern engines don't really do threading that way any more.

Oh, how I wish any of the engines I've worked with in the last 10 years were 'modern'. ;)