Behavior Tree

For my specialization, I chose to work on my own Behavior Tree. The reason behavior trees interest me is that they allow for easy AI setup and debugging, with every behavior simply being a node in the tree's hierarchical structure. To add a behavior, all it takes is adding a node.

I implemented this in our engine, Tonic Engine, which uses its own ECS (Entity Component System), created by William Arnberg (Check portfolio [here]).

A behavior tree follows a hierarchical tree structure, where we program and design each node to represent a specific behavior or action. The execution starts at the root node and activates the child nodes from top to bottom.

It starts at the top (Root node) and activates the child nodes from top to bottom.

The behavior tree consists of a list of nodes, with each node representing a behavior. The root node starts at index 0 and checks whether the tree should be running. Each node has:

• A Node Type, defining the node's role.

• A std::function to store the action or behavior (only used by leaf nodes).

• A vector of children, allowing composite or decorator nodes to be linked to their child nodes.

• Additional variables to handle execution states.

Here are the different types of nodes:

Composite nodes are the ones that control the flow of the execution.

Decorator Nodes allow us to modify the child nodes.

Leaf Nodes perfoms the action or checks to see if we fail or succeed,

Each node has its own logic and functions that we evaluate to determine how it should act based on the results.

Composite

A sequence goes through the children in order, if one fails we return false, if all succed then we return true.

A Selector will run through all the children return true if any of them succed, If all fail then we return false.

Decorator

Inverter inverts the result of its child.

Repeater will run its child a number of times that we set before returning true

Succeeder always runs true, no matter what the result of the child is

Leaf

The End points of our tree, the goal the tree tries to reach.

Condition: Check if a certain condition is met

Action: Perfom a specific action that we either return true or false on.

We can Easily Add our nodes, by giving it the type, sending in the specific entity’s component we want to add it on, a function (specifically for leaf nodes).

Adding a function for other node types should be avoided, same with adding children list for leaf nodes

The update loop starst by sending in the rootnode and only runs if allowed.

The Evaluate Node determines what action it should take, using recursion to evaluate child nodes until a leaf node is reached, then it executes the leaf nodes