参考与代码来自b站-ANVER-大佬

教学视频

以下都是一种固定模板结构，便于外部以及新项目引用。

1.BehaviorTree类

一个BehaviorTree应该包括：
1.Node节点的定义与声明。

2.Blackboard充当大脑用于存储string映射到object的键值对，因为object是所有类型的基类，在转化的过程中会产生很多的拆装箱影响性能，不过由于行为树本身就不是什么庞大（1e6甚至更多那种）底层由哈希表实现在没有哈希冲突的前提下是O（1）的时间复杂度，又由于数据范围很小所以基本不会哈希碰撞,哈希碰撞会导致时间复杂度提高到O(n)。
3.类似状态机结构中的玩家脚本，行为树脚本也是直接挂载在主物体身上的，所以在结构上需要有Blackboard，以及任务或者树的初始状态，类似状态机一开始是处于IdleState，并且在awake的时候获取它和声明它们。
5.在Update里初始化的方法应该是根节点的评估方法。

using JetBrains.Annotations;
using UnityEngine;namespace BehaviourTrees
{[RequireComponent(typeof(Blackboard))]public class BehaviourTree : MonoBehaviour{private Node root;public Node Root{get => root;protected set => root = value;}private Blackboard blackboard;[UsedImplicitly]private void Awake(){blackboard = GetComponent<Blackboard>();OnSetup();}public Blackboard Blackboard{get => blackboard;set => blackboard = value;}[UsedImplicitly]// Update is called once per framevoid Update(){root?.Evaluate(gameObject.transform, blackboard);}protected virtual void OnSetup(){}}
}

2.Node类

一个Node类应该包括：
1.currentState目前的状态{Failure = 0, Success, Running}。
2.该节点的父节点以及该节点的子节点。
3.评估节点状态逻辑函数。

using System;
using System.Collections.Generic;
using UnityEngine;namespace BehaviourTrees
{public enum Status{Failure = 0,Success,Running}public abstract class Node{protected Node parent;protected List<Node> children = new List<Node>();public Status status { get; protected set; }public Status Evaluate(Transform agent, Blackboard blackboard){status = OnEvaluate(agent, blackboard);return status;}protected abstract Status OnEvaluate(Transform agent, Blackboard blackboard);}}

3.composite

复合结构直接继承于Node用于被Sequence和Selector继承

4.Sequence

Sequence：Sequence的逻辑是AND，要求当前队列下的所有子节点都要是Success才返回success否者返回Failure，如果还在执行返回Running。

using System.Collections.Generic;
using System.Linq;
using UnityEngine;namespace BehaviourTrees
{public class Sequencer : Composite{public Sequencer(List<Node> children){this.children = children;}protected override Status OnEvaluate(Transform agent, Blackboard blackboard){bool isRunning = false;bool success = children.All((child) =>{Status status = child.Evaluate(agent, blackboard);switch (status){case Status.Failure:return false;case Status.Running:isRunning = true;break;}return status == Status.Success;});return isRunning ? Status.Running : success ? Status.Success : Status.Failure;}}
}

5.Selector

Selector：一个Selector的逻辑是or,要求当前队列下的所有子节点都要是failure才返回Failure否者返回Success，如果还在执行返回Running。

using System.Collections.Generic;
using System.Linq;
using UnityEngine;namespace BehaviourTrees
{public class Selector : Composite{public Selector(List<Node> children){this.children = children;}protected override Status OnEvaluate(Transform agent, Blackboard blackboard){bool isRunning = false;bool failed = children.All((child) =>{Status status = child.Evaluate(agent, blackboard);if (status == Status.Running) isRunning = true;return status == Status.Failure;});return isRunning ? Status.Running : failed ? Status.Failure : Status.Success;}}
}

6.Task

Task继承于Node应该对Node中的评估方法Evaluate进行重写，用于描述挂载脚本物体的逻辑。

7.Blackboard

using System.Collections;
using System.Collections.Generic;
using Unity.VisualScripting.YamlDotNet.Core.Tokens;
using UnityEngine;namespace BehavioursTree
{public class Blackboard : MonoBehaviour{private Dictionary<string, object> data = new Dictionary<string, object>();public T Get<T>(string key){if (data.TryGetValue(key, out object value)) return (T)value;return default(T);}public void Add<T>(string key, T value){data.Add(key, value);}public bool Remove<T>(string key){if (data.ContainsKey(key)){data.Remove(key);return true;}return false;}}
}

8.实例

①兔子行为树

using BehaviourTrees;
using System.Linq;
using UnityEngine;public class RabbitBehaviourTree : BehaviourTree
{[SerializeField] private Transform[] waypoints = null;[SerializeField] private float speed = 10.0f;protected override void OnSetup(){Blackboard.Add("speed", speed);var patrolTask = new PatrolTask(waypoints);var seeCarrotTask = new SeeCarrotTask();var catchCarrotTask = new CatchCarrotTask();Node[] sequencerChildren = { seeCarrotTask, catchCarrotTask };var sequencer = new Sequencer(sequencerChildren.ToList());Node[] selectorChildren = { sequencer, patrolTask };var selector = new Selector(selectorChildren.ToList());Root = selector;}
}
AWAKEprivate void Awake(){blackboard = GetComponent<Blackboard>();OnSetup();}
UPDATEvoid Update(){root?.Evaluate(gameObject.transform, blackboard);}

这里的兔子行为树定做了：
1.声明巡逻任务。
2.声明探测到萝卜任务。
3.声明吃萝卜任务。
4.声明一个Sequence包含两个子节点①探测到萝卜②吃萝卜。
5.声明一个Seletor包含两个子节点①Sequence②巡逻。

②巡逻任务

using BehaviourTrees;
using UnityEngine;public class PatrolTask : Task
{private int currentIndex;private Transform[] waypoints;public PatrolTask(Transform[] waypoints){this.waypoints = waypoints;currentIndex = 0;}protected override Status OnEvaluate(Transform agent, Blackboard blackboard){float speed = blackboard.Get<float>("speed");Transform currentWaypoint = waypoints[currentIndex];bool arrived = Vector2.Distance(agent.position, currentWaypoint.position) < 0.1f;if (arrived){// update current index++currentIndex;currentIndex %= waypoints.Length;}agent.position = Vector2.MoveTowards(agent.position, currentWaypoint.position, speed * Time.deltaTime);return Status.Running;}
}

巡逻任务包括1：n个巡逻点可以通过Inspector拖动赋值。
2：评估的主要逻辑。

③探测萝卜任务

using System.Collections;
using System.Collections.Generic;
using BehaviourTrees;
using UnityEngine;public class SeeCarrotTask : Task
{private float radius = 2.0f;protected override Status OnEvaluate(Transform agent, Blackboard blackboard){var colliders = Physics2D.OverlapCircleAll(agent.position, radius);if (colliders == null) return Status.Failure;foreach (Collider2D collider in colliders){if (!collider.CompareTag("Carrot")) continue;blackboard.Add("carrot", collider.gameObject);return Status.Success;}return Status.Failure;}
}

探测萝卜任务应该包括：
1.探测半径。

③吃萝卜任务

using System.Collections;
using System.Collections.Generic;
using BehaviourTrees;
using UnityEngine;public class CatchCarrotTask : Task
{protected override Status OnEvaluate(Transform agent, Blackboard blackboard){var carrot = blackboard.Get<GameObject>("carrot");var speed = blackboard.Get<float>("speed");if (carrot == null) return Status.Failure;if (Vector2.Distance(agent.position, carrot.transform.position) <= 0.01f)return Status.Success;Vector2 position = Vector2.MoveTowards(agent.position, carrot.transform.position, speed * Time.deltaTime);position.y = agent.position.y;agent.position = position;return Status.Running;}
}

吃萝卜逻辑：
检测之前在探测萝卜任务中加入到大脑的萝卜object，使用Movetowards进行两点间的移动。

个人对行为树的理解

**行为树的核心机制就是每一帧从根节点开始遍历其子节点，并根据子节点的状态调用对应的 Evaluate 方法。这种行为树的运行方式是典型的深度优先遍历，并且是基于帧的更新（Frame-based Update）。**在这个案例中Selector含有对Sequence和巡逻Task的引用，遍历的顺序就是：注意箭头顺序

Selector->{Sequence{找萝卜 -> 吃萝卜} -> 巡逻} -> 循环