epoch 10

config/ppo/WalkerDynamic.yaml

 behaviors:
   WalkerDynamic:
     trainer: ppo
-    batch_size: 16384 #2048
+    batch_size: 2048
-    buffer_size: 163840 #20480
+    buffer_size: 20480
     epsilon: 0.2
     hidden_units: 512
     lambd: 0.95
     memory_size: 128
     normalize: true
-    num_epoch: 3
+    num_epoch: 10
     num_layers: 3
     time_horizon: 1000
     sequence_length: 64

config/trainer_config.yaml

 WalkerDynamic:
     normalize: true
     # learning_rate_schedule: constant
-    num_epoch: 3
+    num_epoch: 10
+    # batch_size: 24576 #2048
+    # buffer_size: 245760 #20480
+    epsilon: 0.3
     # max_steps: 1e8
     summary_freq: 30000
     # learning_rate: 1.0e-3