multiple seeds

config/sac/3DBall.yaml

       learning_rate: 0.0003
       learning_rate_schedule: constant
       batch_size: 64
-      buffer_size: 12000
+      buffer_size: 500000
       buffer_init_steps: 0
       tau: 0.005
       steps_per_update: 10.0
     network_settings:
       normalize: true
-      hidden_units: 64
+      hidden_units: 128
       num_layers: 2
       vis_encode_type: simple
     reward_signals:

config/sac_transfer/3DBallCloud.yaml

       learning_rate_schedule: linear
       model_schedule: linear
       batch_size: 64
-      buffer_size: 200000
+      buffer_size: 500000
       buffer_init_steps: 0
       tau: 0.005
       steps_per_update: 10.0
-      encoder_layers: 1
-      policy_layers: 1
+      encoder_layers: 2
+      policy_layers: 0
-      feature_size: 16
+      feature_size: 64
       # separate_value_net: true
       separate_policy_train: true
       # separate_value_train: true

config/sac_transfer/3DBallHardCloud.yaml

       save_replay_buffer: false
       init_entcoef: 1.0
       reward_signal_steps_per_update: 10.0
-      encoder_layers: 1
-      policy_layers: 1
+      encoder_layers: 2
+      policy_layers: 0
-      feature_size: 16
+      feature_size: 64
       # separate_value_net: true
       separate_policy_train: true
       # separate_value_train: true

config/sac_transfer/3DBallHardTransferCloud.yaml

       save_replay_buffer: false
       init_entcoef: 1.0
       reward_signal_steps_per_update: 10.0
-      encoder_layers: 1
-      policy_layers: 1
+      encoder_layers: 2
+      policy_layers: 0
-      feature_size: 16
+      feature_size: 64
       # separate_value_net: true
       separate_policy_train: true
       # separate_value_train: true
       train_model: false
       load_policy: true
       load_value: true
-      transfer_path: "results/sac-ball-f16-e1p1f0/3DBall"
+      transfer_path: "results/ball-linear-s0/3DBall"
     network_settings:
       normalize: true
       hidden_units: 128