A brief introduction to the behavior tree of unity AI

1. Behavior tree birth

Designing games AI When , Our goal is to find a simple , The scheme of extensible editing logic , So as to accelerate the iteration speed of game development .

Take soldiers as an example , Suppose soldiers have Free 、 fighting 、 escape Three states , State machine （FSM） Is the first plan to come to mind , Then the state machine diagram of soldiers can be shown as follows .

But as the development goes on , State one more , State machine maintenance is not so easy , State machine The transition line between them is like a runaway horse , Can't control . For example, add another hide and return to the starting state , Then the connection will become more and more complex .

So the predecessors designed Behavior tree Abstract to solve this problem .

2. What is a behavior tree ？

Behavior tree , English is Behavior Tree, abbreviation BT, It's a tree for controlling AI Decision making behavior 、 A tree structure containing hierarchical nodes .

2.1 principle ： Traverse

When we have to decide what kind of behavior this soldier is going to do , We'll be top-down , Search the tree by some criteria , Finalize what needs to be done （ Leaf nodes ）, And execute it , This is the basic principle of behavior tree .

2.2 form

The behavior tree is mainly abstracted from the following four nodes Combination of node 、 Decoration node 、 Conditional node 、 Behavior node .

① Combination of node （Composites）

It mainly includes ：Sequence Sequence condition ,Selector Choose the conditions ,Parallel Parallel conditions and the conditions of their combination .

② Decorated nodes （Decorator）

Branches connecting leaves , It is all kinds of decoration nodes , These nodes determine AI How to start from the top of the tree according to different situations , To follow different paths to the final leaf .
For example, let the child nodes operate circularly （LOOP） Or let the son task Run until it returns a certain running state value （Util）, Or will task Negative return value of （NOT） wait

③ Conditional node （Conditinals）

Used to judge whether a condition is true . So far , yes Behavior Designer In order to implement the principle of single responsibility , Treat the judgment as a node independently , For example, judge whether a target is in the field of vision , In fact, they are attacking Action It can also be written in it , But this way Action It's not single , It is not conducive to the reuse of visual field judgment processing . The general condition node appears in Sequence In the control node , Followed by Action node .

④ Behavior node （Action）

Behavior node is the node that really does things , The behavior node is at the end of the tree , It's all leaf nodes , That's all. AI Orders to actually do things ;

Drive the behavior of the character by using the association between the nodes in the behavior tree , Rather than telling a character what to do with specific code , It's much more interesting to come , This is also the most exciting point of the behavior tree . In this way, we only need to abstract good behavior , Don't care what happened in the battle .

3. Tree traversal -Tree Traversal

One characteristic of the behavior tree is , It will “ One layer at a time ” Check the nodes in turn , And each floor needs to spend one tick Time for , So it takes several tick To complete the process from the top to the bottom , To complete its logic , This is very different from using code to realize functions .

This is not a very efficient way , Especially when your tree gets very deep . I think the implementation of behavior tree must have the ability to be in a tick Complete the judgment logic of the whole behavior tree .

4. workflow -Flow

The behavior tree consists of many different types of nodes , They all have a common core function , That is, they will return one of the three states as a result . These three states are ：

• success -Success;
• Failure -Failure;
• Running -Running;

The first two , Just like their name , It is used to inform their parent node of the success or failure of the operation . The third means that it is still running , The result has not been decided yet , The next tick Check the operation results of this node when you need to .

This function is very important , It can keep a node running for a period of time to maintain some behavior .

For example, a “walk（ walk ）” The node will continue to return during the process of calculating pathfinding and keeping the character walking “Running” To keep the character in this state .
If pathfinding fails for some reason , Or except for some conditions that make the walking behavior have to stop , Then this node will return “Failure” To tell its parent node ;
If the character goes to the designated destination , Then the node returns “Success” To indicate that the walking command has been successfully completed .

These states can be used to determine the direction of the behavior tree , Make sure AI The behaviors in the behavior tree can be executed in some order in the way we expect .

See the article at the end of the article for specific introduction , Here is just a record and introduction

Data context -Data Context

When a behavior tree is called , A data context is also created .

Its purpose is to store variables interpreted and changed by nodes . These variables can then be read and written by the node in combination with the context of the data , So that the whole behavior tree remains a unified whole .

Behavior tree description ：

Behavior trees are various classic control nodes + Combine behavior nodes , So as to realize the control of some complex behavior states , The behavior tree plug-in used here is Behavior Designer, It contains four types of nodes . Namely action node 、 Combination of node 、 Condition node and decoration node .
(1) Composites Combination of node , It mainly includes ：Sequence Sequence condition ,Selector Choose the conditions ,Parallel Parallel conditions and the conditions of their combination .
(2) Decorator Decoration node , This node contains only one child node , There are no special functions added to this child node , For example, let child nodes operate circularly or let children task Run until it returns a certain running state value , Or will task And so on
(3) Actions Behavior node , Behavior node is the node that really does things , It is a leaf node .Behavior Designer The plug-in comes with many Action node , If not enough , You can also write your own Action. For the behavior state written by yourself , Generally, you need to add your own
(4) Conditinals Conditional node , Used to judge whether a condition is true . So far , yes Behavior Designer In order to implement the principle of single responsibility , Treat the judgment as a node independently , For example, judge whether a target is in the field of vision , In fact, they are attacking Action It can also be written in it , But this way Action It's not single , It is not conducive to the reuse of visual field judgment processing . The general condition node appears in Sequence In the control node , Followed by Action node .
Behavior Designer The sharing of variables is handled as follows ：
In the same Behavior Tree（ Generally one GameObject There is one Behavior Tree） Of Task Local variables shared between can be directly in the editor Variables add to ; In addition, it also supports in different Behavior Tree Global variables shared between ; And support Task And non Task（ Other scripts in the game system ） Pass variables between , Through the following code ：
behaviorTree.GetVariableName(“MyVariableName”);
behaviorTree.SetVariableName(“MyVariableName”, value);

Write your own behavior subclasses , You can create nodes with customized behavior , These nodes need to inherit one of the four nodes . Such as , We can create a condition node for monster patrol , That is, the monster defaults to patrol , When the distance between the player and the monster is less than a certain distance , Monsters start chasing players .

Reference link ：

Wu Yifeng ： game AI - Behavior tree Part1： brief introduction

Wu Yifeng ： game AI - Behavior tree Part2： frame

AI How behavior trees work | indienova Independent game