opponent observations

Project/Assets/ML-Agents/Examples/Tennis/Prefabs/TennisArea.prefab

   m_Name: 
   m_EditorClassIdentifier: 
   m_BrainParameters:
-    vectorObservationSize: 10
+    vectorObservationSize: 14
-  m_Model: {fileID: 11400000, guid: d6c5e749e4ceb4cf79640a5955706d3d, type: 3}
+  m_Model: {fileID: 11400000, guid: 13ca629afc16c42afb2d681a3d81ea6c, type: 3}
   m_InferenceDevice: 0
   m_BehaviorType: 0
   m_BehaviorName: Tennis
   hasUpgradedFromAgentParameters: 1
   maxStep: 3000
   ball: {fileID: 1273406647218856}
+  opponent: {fileID: 1882383181950958}
-  timePenalty: 0
+  energyPenalty: 0
 --- !u!114 &2449890524009497851
 MonoBehaviour:
   m_ObjectHideFlags: 0
   m_Name: 
   m_EditorClassIdentifier: 
   m_BrainParameters:
-    vectorObservationSize: 10
+    vectorObservationSize: 14
-  m_Model: {fileID: 11400000, guid: d6c5e749e4ceb4cf79640a5955706d3d, type: 3}
+  m_Model: {fileID: 11400000, guid: 13ca629afc16c42afb2d681a3d81ea6c, type: 3}
   m_InferenceDevice: 0
   m_BehaviorType: 0
   m_BehaviorName: Tennis
   hasUpgradedFromAgentParameters: 1
   maxStep: 3000
   ball: {fileID: 1273406647218856}
+  opponent: {fileID: 1170495812642400}
-  timePenalty: 0
+  energyPenalty: 0
 --- !u!114 &6598495797138489682
 MonoBehaviour:
   m_ObjectHideFlags: 0

Project/Assets/ML-Agents/Examples/Tennis/Scripts/TennisAgent.cs

 {
     [Header("Specific to Tennis")]
     public GameObject ball;
+    public GameObject opponent;
     public bool invertX;
     public int score;
     public GameObject myArea;
     Text m_TextComponent;
     Rigidbody m_AgentRb;
     Rigidbody m_BallRb;
+    Rigidbody m_OpponentRb;
     HitWall m_BallScript;
     TennisArea m_Area;
     float m_InvertMult;
     {
         m_AgentRb = GetComponent<Rigidbody>();
         m_BallRb = ball.GetComponent<Rigidbody>();
+        m_OpponentRb = opponent.GetComponent<Rigidbody>();
         m_BallScript = ball.GetComponent<HitWall>();
         m_Area = myArea.GetComponent<TennisArea>();
         var canvas = GameObject.Find(k_CanvasName);
         sensor.AddObservation(m_InvertMult * m_BallRb.velocity.x);
         sensor.AddObservation(m_BallRb.velocity.y);
 
+        sensor.AddObservation(m_InvertMult * (opponent.transform.position.x - myArea.transform.position.x));
+        sensor.AddObservation(opponent.transform.position.y - myArea.transform.position.y);
+        sensor.AddObservation(m_InvertMult * m_OpponentRb.velocity.x);
+        sensor.AddObservation(m_OpponentRb.velocity.y);
+
-        AddReward(m_BallTouch * (1f /  Vector3.Distance(ball.transform.position, transform.position)));
     }
 
     public override void OnActionReceived(float[] vectorAction)

config/trainer_config.yaml

 Tennis:
     normalize: true
     max_steps: 5.0e7
-    learning_rate: 3.0e-4
-    hidden_units: 256
+    hidden_units: 512
     beta: 1.0e-2
     time_horizon: 1000
     self_play: