push block more steps

Project/Assets/ML-Agents/Examples/Reacher/Scripts/NewReacherAgent.cs

         var torqueX = Mathf.Clamp(vectorAction[0], -1f, 1f) * 150f;
         var torqueZ = Mathf.Clamp(vectorAction[1], -1f, 1f) * 150f;
         m_RbA.AddTorque(new Vector3(torqueX, 0f, torqueZ));
-
-        AddReward( - (0.05f * torqueX * torqueX + 0.05f * torqueZ * torqueZ));
-
+        
-        AddReward( - (0.05f * torqueX * torqueX + 0.05f * torqueZ * torqueZ));
+        AddReward( - 0.005f * (vectorAction[0] * vectorAction[0] 
+                        + vectorAction[1] * vectorAction[1] 
+                        + vectorAction[2] * vectorAction[2] 
+                        + vectorAction[3] * vectorAction[3] 
+                ));
     }
 
     /// <summary>

Project/Assets/ML-Agents/Examples/Reacher/Scripts/ReacherAgent.cs

 
         sensor.AddObservation(goal.transform.localPosition);
         sensor.AddObservation(hand.transform.localPosition);
-        
+
         sensor.AddObservation(m_GoalSpeed);
     }
 
         var torqueZ = Mathf.Clamp(vectorAction[1], -1f, 1f) * 150f;
         m_RbA.AddTorque(new Vector3(torqueX, 0f, torqueZ));
 
-        AddReward( - (0.05f * torqueX * torqueX + 0.05f * torqueZ * torqueZ));
+        
-        AddReward( - (0.05f * torqueX * torqueX + 0.05f * torqueZ * torqueZ));
+        AddReward( - 0.005f * (vectorAction[0] * vectorAction[0] 
+                        + vectorAction[1] * vectorAction[1] 
+                        + vectorAction[2] * vectorAction[2] 
+                        + vectorAction[3] * vectorAction[3] 
+                ));
     }
 
     /// <summary>

config/sac_transfer/PushBlock.yaml

       learning_rate_schedule: linear
       model_schedule: constant
       batch_size: 128
-      buffer_size: 2000000
+      buffer_size: 6000000
       buffer_init_steps: 0
       tau: 0.005
       steps_per_update: 10.0
       forward_layers: 2
       value_layers: 2
       action_layers: 2
-      feature_size: 256
+      feature_size: 128
       action_feature_size: 64
       separate_policy_train: true
       separate_policy_net: true
         gamma: 0.99
         strength: 1.0
     keep_checkpoints: 5
-    max_steps: 2000000
+    max_steps: 6000000
     time_horizon: 64
     summary_freq: 100000
     threaded: true

config/sac_transfer/PushBlockTransfer.yaml

       train_model: false
       load_action: true
       train_action: false
-      transfer_path: "results/block-f256/PushBlock"
+      transfer_path: "results/block-long/PushBlock"
     network_settings:
       normalize: false
       hidden_units: 256