Goal-Oriented Action Planners

[0ffz]

Goal-Oriented Action Planning (GOAP) is a technique that essentially builds behavior trees on its own, given the actions it can take. This is far better suited to a dynamic game like Minecraft, and the actions are also quite well suited to an ECS. GOAP also fits nicely with Minecraft's Pathfinder Goal system.

The system

• A goal can be defined with a priority, alongside any other PathfinderGoals. The highest priority goal which can be executed will be executed.
• Actions specify what they want to do. More specifically, what the conditions are to get that thing done, and what the results are (we call these pre/postconditions). Actions also have a "cost" associated with doing them.
• Goals also specify a condition they want to see met.
• When seeing if a goal can be executed, we pathfind backwards from the goal, going through actions with the lowest cost first, and seeing whether their conditions can be met, connecting them to actions which cause those conditions to occur. Essentially we are trying to find the lowest cost path to some action which has its conditions met, and which will, perhaps through several actions, eventually meet the conditions we define in our goal. Simple A* pathfinding can be used here.

Resources

A nice overview of common Behavior Selection Algorithms:
http://www.gameaipro.com/GameAIPro/GameAIPro_Chapter04_Behavior_Selection_Algorithms.pdf