Behavior Trees

Behavior trees are less formal than finite state machines, they are more of a software engineering tool that expands on finite state machines.

They are setup like a tree, the root node is the highest level goal and you traverse leafs.

Many game engines have replaced state machines with behavior tree implementations.

Ex: pyTrees.

Sequence Nodes

A sequence node has multiple sequential leaf nodes. You walk the sequence, and if one leaf node fails the whole sequence node goes to a failed state. If they all succeed the sequence node succeeds.

Selector Nodes

Executes leaves from left to right until the first one succeeds - try different things in decreasing priority.

Parallel Node

Execute tasks in parallel, only succeed if all children succeed.

Decorators

Nodes with special behavior that apply to any child or children, such as "repeat".

Blackboards

• Shared memory
• Controlled read and write behavior

Real-time Execution Model

• Introduce new state RUNNING
• Return to last RUNNING state after each tick
• We can now implement real-time reactive behaviors

Advantages

• Forces engineers to consider failures early in the process
• Composability (because SUCCESS, FAILURE, and RUNNING gives a high level general interface for all nodes)