Behavior trees and state machines with LimboAI

LimboAI addon overview

We'll be using LimboAI for behavior trees and state machines

LimboAI docs | LimboAI on GitHub

LimboAI includes a demo project. It is highly recommended that you download it and check it out. When you run it, it has a nice little tutorial that walks you through the basics: https://github.com/limbonaut/limboai/tree/master/demo 

In its classes and nodes, LimboAI uses the following prefixes:

• BT: Behavior Tree
  • Nodes/classes that start with this have to do with parts of behavior trees
  • As an example, the BTPlayer node "plays" a behavior tree every frame and inside of that tree you have BTSequence, BTSelector, etc
• BB: BlackBoard
  • The blackboard is a shared place for agents to write values to.
  • These values can be read and written by any task in the tree.
• LimboHSM: Hierarchical State Machine
  • HSMs have LimboStates inside of them.
  • LimboHSM is itself a LimboState, and BTState can put states inside of behavior trees

Behavior trees

Read the intro to behavior trees here

Accessing LimboAI behavior tree editor

When installed, LimboAI adds a button to the top of the Godot editor, alongside the 2D/3D/Script buttons

When you click on a BTPlayer node, it has a BehaviorTree attached to it. When clicking that, it will automatically switch over to the LimboAI tab.

Building blocks: sequences and selectors

The main building blocks of a behavior tree are:

• BTSequence: Think of these like an AND operator on the child tasks.
  • As long as a task return SUCCESS, the next one will be executed.
• BTSelector: Think of these like an OR operator on the child tasks.
  • On the first task that returns SUCCESS, it will stop executing.

Writing custom tasks

Read the documentation on writing custom tasks.

Custom tasks go inside of ai/tasks in the project.

At the time of writing, we have:

• follow_path: Follows a Path3D node (i.e. for preset patrol paths)
• generate_navigation_path / follow_navigation_path: Using a NavigationAgent3D, generate and follow a path to a target in the blackboard
  • follow_navigation_path will follow the path until it gets close enough to the target, it will return RUNNING until it gets there
• distance_based_cooldown: Takes in a Curve2D and uses that to do a cooldown based on how close the agent is to the target
  • Useful for generating navigation paths more frequently the closer you are to the target, and less frequently the further you are
• in_shape_and_visible: Returns SUCCESS if there's a target in a ShapeCast3D that is visible (hit by a raycast, in "front" of the agent)
  • This also sets the target when it is visible

State machines

https://limboai.readthedocs.io/en/stable/hierarchical-state-machines/create-hsm.html