# 體驗 ML-Agent (RollerBall)

簡單的滾球吃目標
目標:碰到綠色方塊
代理:一個紅球

首先設置環境和代理
放上地板(設定範圍)
加上目標Target(綠色方塊)和代理Agent(紅球)

開始設計規則和Agent的行動

public class RollerAgent : Agent
{
    private Rigidbody rBody;
    public Transform target;

    // Start is called before the first frame update
    void Start()
    {
        rBody = GetComponent<Rigidbody>();
    }

    public override void OnEpisodeBegin()
    {
        if(this.transform.localPosition.y < 0)
        {
            this.rBody.velocity = Vector3.zero;
            this.rBody.angularVelocity = Vector3.zero;
            this.transform.localPosition = new Vector3(0, 0.5f, 0);
        }
        target.localPosition = new Vector3(Random.Range(-4f, 4f), 0.5f, Random.Range(-4f, 4f));
    }

    public override void CollectObservations(VectorSensor sensor) //需要觀測agent或是target行為的點  //8個
    {
        sensor.AddObservation(target.localPosition);                                                 //3
        sensor.AddObservation(this.transform.localPosition);                                         //3

        sensor.AddObservation(rBody.velocity.x);                                                     //1
        sensor.AddObservation(rBody.velocity.z);                                                     //1
    }

    public float forceMult = 10f;
    public override void OnActionReceived(ActionBuffers actions) //動作和給予獎勵
    {
        float x = actions.ContinuousActions[0];
        float z = actions.ContinuousActions[1];

        rBody.AddForce(new Vector3(x, 0, z) * forceMult);

        float dist = Vector3.Distance(this.transform.localPosition, target.localPosition);

        if (dist < 1.42f)
        {
            SetReward(1.0f);
            EndEpisode();
        }

        else if (this.transform.localPosition.y < 0)
        {
            EndEpisode();
        }
    }

    public override void Heuristic(in ActionBuffers actionsOut)  //keyboard控制test
    {
        var continuousActionsOut = actionsOut.ContinuousActions;
        continuousActionsOut[0] = Input.GetAxis("Horizontal");
        continuousActionsOut[1] = Input.GetAxis("Vertical");
    }
}


調整好後，就將全部複製以方便訓練


訓練參數設置:

behaviors:
  RollerBall:
    trainer_type: ppo
    hyperparameters:
      batch_size: 10
      buffer_size: 100
      learning_rate: 3.0e-4
      beta: 5.0e-4
      epsilon: 0.2
      lambd: 0.99
      num_epoch: 3
      learning_rate_schedule: linear
      beta_schedule: constant
      epsilon_schedule: linear
    network_settings:
      normalize: false
      hidden_units: 128
      num_layers: 2
    reward_signals:
      extrinsic:
        gamma: 0.99
        strength: 1.0
    max_steps: 50000
    time_horizon: 64
    summary_freq: 10000

以上都完成，就可以開啟在Anaconda設置好的環境來進行訓練了

mlagents-learn config/rollerball.yaml --run-id=RollerBall

成功出現Unity圖型就表示參數呼叫成功，回到Unity按下開始就會開始訓練了

訓練完成

成品:
{%youtube mWf2kcA2kqE %}

使用GPU