no normalization

config/ppo_transfer/CrawlerStaticOpbuffer.yaml

     hyperparameters:
       batch_size: 2024
       buffer_size: 20240
-      learning_rate: 0.0003
+      learning_rate: 0.0002
       beta: 0.005
       epsilon: 0.2
       lambd: 0.95
       feature_size: 32
       reuse_encoder: true
       in_epoch_alter: true
+      use_op_buffer: true
+      separate_value_net: true
+      use_var_predict: true
+      with_prior: true
     network_settings:
       normalize: true
       hidden_units: 512

config/ppo_transfer/WalkerStaticSingle.yaml

     hyperparameters:
       batch_size: 2048
       buffer_size: 20480
-      learning_rate: 0.0002
+      learning_rate: 0.0005
       beta: 0.005
       epsilon: 0.2
       lambd: 0.95
       forward_layers: 2
       inverse_layers: 1
       separate_value_net: true
-      feature_size: 64
+      feature_size: 128
       reuse_encoder: true
       in_epoch_alter: true
       use_op_buffer: true

ml-agents/mlagents/trainers/policy/transfer_policy.py

             self.brain.camera_resolutions
         )
         self.vector_next = ModelUtils.create_vector_input(self.vec_obs_size)
-        if self.normalize:
-            self.processed_vector_next = ModelUtils.normalize_vector_obs(
-                self.vector_next,
-                self.running_mean,
-                self.running_variance,
-                self.normalization_steps,
-            )
-        else:
-            self.processed_vector_next = self.vector_next
+        # if self.normalize:
+        #     self.processed_vector_next = ModelUtils.normalize_vector_obs(
+        #         self.vector_next,
+        #         self.running_mean,
+        #         self.running_variance,
+        #         self.normalization_steps,
+        #     )
+        # else:
+        #     self.processed_vector_next = self.vector_next
-                self.processed_vector_next,
+                self.vector_next,
                 1,
                 h_size,
                 num_layers,
             self.brain.camera_resolutions
         )
         self.vector_next = ModelUtils.create_vector_input(self.vec_obs_size)
-        if self.normalize:
-            self.processed_vector_next = ModelUtils.normalize_vector_obs(
-                self.vector_next,
-                self.running_mean,
-                self.running_variance,
-                self.normalization_steps,
-            )
-        else:
-            self.processed_vector_next = self.vector_next
+        # if self.normalize:
+        #     self.processed_vector_next = ModelUtils.normalize_vector_obs(
+        #         self.vector_next,
+        #         self.running_mean,
+        #         self.running_variance,
+        #         self.normalization_steps,
+        #     )
+        # else:
+        #     self.processed_vector_next = self.vector_next
-                self.processed_vector_next,
+                self.vector_next,
                 1,
                 h_size,
                 num_layers,

ml-agents/mlagents/trainers/ppo_transfer/optimizer.py

             self.policy.mask_input: mini_batch["masks"] * burn_in_mask,
             self.advantage: mini_batch["advantages"],
             self.all_old_log_probs: mini_batch["action_probs"],
-            self.policy.processed_vector_next: mini_batch["next_vector_in"],
+            self.policy.vector_next: mini_batch["next_vector_in"],
             # self.policy.next_vector_in: mini_batch["next_vector_in"],
             self.policy.current_action: mini_batch["actions"],
             self.policy.current_reward: mini_batch["extrinsic_rewards"],

ml-agents/mlagents/trainers/tests/test_simple_transfer.py

         learning_rate_schedule=ScheduleType.CONSTANT,
         batch_size=16,
         buffer_size=64,
-        feature_size=2,
+        feature_size=4,
         reuse_encoder=True,
         in_epoch_alter=True,
         # in_batch_alter=True,
     #     assert all(reward > success_threshold for reward in processed_rewards)
 
 
-def test_2d_model(config=Transfer_CONFIG, obs_spec_type="normal", run_id="model_normal", seed=0):
+def test_2d_model(config=Transfer_CONFIG, obs_spec_type="normal", run_id="model_normal_f4_varp-pri-test", seed=0):
-        config.hyperparameters, batch_size=120, buffer_size=12000, learning_rate=5.0e-3
+        config.hyperparameters, batch_size=120, buffer_size=12000, learning_rate=5.0e-3,
+        use_var_predict=True, with_prior=True
-def test_2d_transfer(config=Transfer_CONFIG, obs_spec_type="rich2", run_id="transfer_rich2_from-rich1", seed=1337):
+def test_2d_transfer(config=Transfer_CONFIG, obs_spec_type="rich1", run_id="transfer_rich1_from-normal_varp-pri_retrain-all_5e-2", seed=1337):
     env = SimpleTransferEnvironment(
         [BRAIN_NAME], use_discrete=False, action_size=2, step_size=0.1, 
         num_vector=2, obs_spec_type=obs_spec_type, goal_type="hard"
-        transfer_path="./transfer_results/model_rich1_s0/Simple",
-        use_op_buffer=True, in_epoch_alter=True, in_batch_alter=False, learning_rate=5e-4, 
-        train_policy=False, train_value=False, train_model=False, feature_size=2
+        transfer_path="./transfer_results/model_normal_f4_varp-pri_s0/Simple",
+        use_op_buffer=True, in_epoch_alter=True, in_batch_alter=False, learning_rate=5.0e-2, 
+        train_policy=True, train_value=True, train_model=True, feature_size=4, learning_rate_schedule=ScheduleType.LINEAR,
+        use_var_predict=True, with_prior=True
     )
     config = attr.evolve(config, hyperparameters=new_hyperparams, max_steps=200000, summary_freq=5000)
     _check_environment_trains(env, {BRAIN_NAME: config}, run_id=run_id + "_s" + str(seed), seed=seed)
     # test_2d_model(seed=0)
     # test_2d_model(config=PPO_CONFIG, run_id="ppo_normal", seed=0)
-    test_2d_transfer(seed=0)
+    test_2d_transfer(seed=123)
     # for i in range(5):
     #     test_2d_model(seed=i)