more experiments for self-play

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

 {
     public GameObject areaObject;
     public int lastAgentHit;
-    public bool net;
 
     public enum FloorHit
         {
         m_AgentB.EndEpisode();
         m_Area.MatchReset();
         lastFloorHit = FloorHit.Service;
-        net = false;
+        lastAgentHit = -1;
     }
     
     void AgentAWins()
                 else
                 {
                     lastFloorHit = FloorHit.FloorAHit;
-                    //successful serve
-                    if (!net)
-                    {
-                        net = true;
-                    }
                 }
             }
             else if (collision.gameObject.name == "floorB")
                 else
                 {
                     lastFloorHit = FloorHit.FloorBHit;
-                    //successful serve
-                    if (!net)
-                    {
-                        net = true;
-                    }
-                }
-            }
-            else if (collision.gameObject.name == "net" && !net)
-            {
-                if (lastAgentHit == 0)
-                {
-                    AgentBWins();
-                }
-                else if (lastAgentHit == 1)
-                {
-                    AgentAWins();
                 }
             }
         }
             }
             else
             {
-                //agent can return serve in the air
-                if (lastFloorHit != FloorHit.Service && !net)
-                {
-                    net = true;
-                }
 
                 lastAgentHit = 0;
                 lastFloorHit = FloorHit.FloorHitUnset;
             }
             else
             {
-                if (lastFloorHit != FloorHit.Service && !net)
-                {
-                    net = true;
-                }
-
                 lastAgentHit = 1;
                 lastFloorHit = FloorHit.FloorHitUnset;
             }

config/trainer_config.yaml

 
 Goalie:
     normalize: false
-    max_steps: 3.0e7
+    max_steps: 5.0e7
     learning_rate_schedule: constant
     batch_size: 2048
     buffer_size: 20480
 
 Striker:
     normalize: false
-    max_steps: 3.0e7
+    max_steps: 5.0e7
     learning_rate_schedule: constant
     batch_size: 2048
     buffer_size: 20480
 
 SoccerTwos:
     normalize: false
-    max_steps: 5.0e7
+    max_steps: 1.0e8
     learning_rate_schedule: constant
     batch_size: 2048
     buffer_size: 20480