reacher: stack observation. with the original reward function

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

         torqueZ = Mathf.Clamp(vectorAction[3], -1f, 1f) * 150f;
         m_RbB.AddTorque(new Vector3(torqueX, 0f, torqueZ));
 
-        AddReward( - 0.001f * (vectorAction[0] * vectorAction[0] 
-                + vectorAction[1] * vectorAction[1] 
-                + vectorAction[2] * vectorAction[2] 
-                + vectorAction[3] * vectorAction[3] 
-        ));
+        // AddReward( - 0.001f * (vectorAction[0] * vectorAction[0] 
+        //         + vectorAction[1] * vectorAction[1] 
+        //         + vectorAction[2] * vectorAction[2] 
+        //         + vectorAction[3] * vectorAction[3] 
+        // ));
     }
 
     /// <summary>
     {
-        if ((goal.transform.position - hand.transform.position).magnitude > 3.5f)
-        {
-            AddReward(-0.002f);
-        }
+        // if ((goal.transform.position - hand.transform.position).magnitude > 3.5f)
+        // {
+        //     AddReward(-0.002f);
+        // }
         // AddReward( - 0.001f * (goal.transform.position - hand.transform.position).magnitude);
         // Debug.Log((goal.transform.position - hand.transform.position).magnitude);
         var radians = m_GoalDegree * Mathf.PI / 180f;

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

         torqueZ = Mathf.Clamp(vectorAction[3], -1f, 1f) * 150f;
         m_RbB.AddTorque(new Vector3(torqueX, 0f, torqueZ));
 
-        AddReward( - 0.001f * (vectorAction[0] * vectorAction[0] 
-                + vectorAction[1] * vectorAction[1] 
-                + vectorAction[2] * vectorAction[2] 
-                + vectorAction[3] * vectorAction[3] 
-        ));
+        // AddReward( - 0.001f * (vectorAction[0] * vectorAction[0] 
+        //         + vectorAction[1] * vectorAction[1] 
+        //         + vectorAction[2] * vectorAction[2] 
+        //         + vectorAction[3] * vectorAction[3] 
+        // ));
     }
 
     /// <summary>
     {
-        if ((goal.transform.position - hand.transform.position).magnitude > 3.5f)
-        {
-            AddReward(-0.002f);
-        }
+        // if ((goal.transform.position - hand.transform.position).magnitude > 3.5f)
+        // {
+        //     AddReward(-0.002f);
+        // }
         // AddReward( - 0.001f * (goal.transform.position - hand.transform.position).magnitude);
         // Debug.Log((goal.transform.position - hand.transform.position).magnitude);
         var radians = m_GoalDegree * Mathf.PI / 180f;

config/sac_transfer/Reacher.yaml

       learning_rate_schedule: linear
       model_schedule: constant
       batch_size: 256
-      buffer_size: 6000000
+      buffer_size: 10000000
       buffer_init_steps: 0
       tau: 0.005
       steps_per_update: 20.0
         gamma: 0.99
         strength: 1.0
     keep_checkpoints: 5
-    max_steps: 6000000
+    max_steps: 10000000
     time_horizon: 1000
     summary_freq: 60000
     threaded: true

config/sac_transfer/ReacherTransfer.yaml

   Reacher:
     trainer_type: sac_transfer
     hyperparameters:
-      learning_rate: 0.0006
-      learning_rate_schedule: constant
+      learning_rate: 0.0003
+      learning_rate_schedule: linear
       model_schedule: constant
       batch_size: 256
       buffer_size: 6000000
       train_model: false
       load_action: true
       train_action: false
-      transfer_path: "results/reacher/Reacher"
+      transfer_path: "results/reacher-ori-sta/Reacher"
     network_settings:
       normalize: true
       hidden_units: 128

config/sac_transfer/3DBall1fTransfer.yaml

+behaviors:
+  3DBall:
+    trainer_type: sac_transfer
+    hyperparameters:
+      learning_rate: 0.0003
+      learning_rate_schedule: linear
+      model_schedule: constant
+      batch_size: 64
+      buffer_size: 500000
+      buffer_init_steps: 0
+      tau: 0.005
+      steps_per_update: 10.0
+      save_replay_buffer: false
+      init_entcoef: 0.5
+      reward_signal_steps_per_update: 10.0
+      encoder_layers: 2
+      policy_layers: 0
+      forward_layers: 0
+      value_layers: 1
+      action_layers: 1
+      feature_size: 64
+      action_feature_size: 32
+      # separate_value_net: true
+      separate_policy_train: true
+      # separate_value_train: true
+      separate_model_train: true
+      reuse_encoder: true
+      in_epoch_alter: false
+      in_batch_alter: true
+      use_op_buffer: false
+      use_var_predict: true
+      with_prior: false
+      predict_return: true
+      use_bisim: false
+      use_transfer: true
+      load_model: false
+      load_encoder: true
+      train_encoder: false
+      load_action: true
+      train_action: false
+      transfer_path: "results/ball_1f/3DBall"
+    network_settings:
+      normalize: true
+      hidden_units: 128
+      num_layers: 2
+      vis_encode_type: simple
+    reward_signals:
+      extrinsic:
+        gamma: 0.99
+        strength: 1.0
+    keep_checkpoints: 5
+    max_steps: 500000
+    time_horizon: 1000
+    summary_freq: 12000
+    threaded: true