Merge branch 'benchmark' of github.com:middleyuan/safe-control-gym in…

…to benchmark

examples/rl/config_overrides/quadrotor_2D_attitude/dppo_quadrotor_2D_attitude.yaml

@@ -1,57 +1,58 @@
-algo_config:
-  # model args
-  hidden_dim: 128
-  activation: "tanh"
+ algo_config:
+   # model args
+   hidden_dim: 128
+   activation: "tanh"
 
-  # loss args
-  gamma: 0.98
-  use_gae: True
-  gae_lambda: 0.92
-  clip_param: 0.2
-  target_kl: 1.0e-2
-  entropy_coef: 0.01
-  quantile_count: 256
-  value_loss: 'quantile_l1'
+   # loss args
+   gamma: 0.98
+   use_gae: True
+   gae_lambda: 0.92
+   clip_param: 0.2
+   target_kl: 1.0e-2
+   entropy_coef: 0.005
+   quantile_count: 256
+   value_loss: 'quantile_l1'
 
-  # optim args
-  opt_epochs: 20
-  mini_batch_size: 256
-  actor_lr: 0.001
-  critic_lr: 0.001
+   # optim args
+   opt_epochs: 20
+   mini_batch_size: 256
+   actor_lr: 0.001
+   critic_lr: 0.001
 
-  # runner args
-  max_env_steps: 396000
-  rollout_batch_size: 5
-  rollout_steps: 660
-  eval_batch_size: 10
+   # runner args
+   max_env_steps: 660000
+   rollout_batch_size: 5
+   rollout_steps: 660
+   eval_batch_size: 10
 
-  # misc
-  log_interval: 6600
-  save_interval: 660000
-  num_checkpoints: 0
-  eval_interval: 6600
-  eval_save_best: True
-  tensorboard: False
+   # misc
+   log_interval: 13200
+   save_interval: 660000
+   num_checkpoints: 0
+   eval_interval: 13200
+   eval_save_best: True
+   tensorboard: False
 
 #algo_config:
 #  # model args
-#  hidden_dim: 128
-#  activation: "relu"
+#  hidden_dim: 8
+#  activation: "tanh"
 #
 #  # loss args
 #  gamma: 0.98
-#  use_gae: True   # or False
-#  gae_lambda: 0.92
-#  use_clipped_value: False  # or True
-#  clip_param: 0.1
-#  target_kl: 1.0e-5
-#  entropy_coef: 0.005
+#  use_gae: True
+#  gae_lambda: 0.9
+#  clip_param: 0.2
+#  target_kl: 2.32e-2
+#  entropy_coef: 0.09
+#  quantile_count: 256
+#  value_loss: 'quantile_l1'
 #
 #  # optim args
 #  opt_epochs: 20
 #  mini_batch_size: 256
-#  actor_lr: 0.0005
-#  critic_lr: 0.0005
+#  actor_lr: 0.0012
+#  critic_lr: 0.0012
 #
 #  # runner args
 #  max_env_steps: 660000

examples/rl/config_overrides/quadrotor_2D_attitude/ppo_quadrotor_2D_attitude.yaml

@@ -1,66 +1,65 @@
-algo_config:
-  # model args
-  hidden_dim: 128
-  activation: "tanh"
+ algo_config:
+   # model args
+   hidden_dim: 128
+   activation: "tanh"
 
-  # loss args
-  gamma: 0.98
-  use_gae: True
-  gae_lambda: 0.92
-  clip_param: 0.2
-  target_kl: 1.0e-2
-  entropy_coef: 0.01
+   # loss args
+   gamma: 0.98
+   use_gae: True
+   gae_lambda: 0.92
+   clip_param: 0.2
+   target_kl: 1.0e-2
+   entropy_coef: 0.005
 
-  # optim args
-  opt_epochs: 20
-  mini_batch_size: 256
-  actor_lr: 0.001
-  critic_lr: 0.001
+   # optim args
+   opt_epochs: 20
+   mini_batch_size: 256
+   actor_lr: 0.001
+   critic_lr: 0.001
 
-  # runner args
-  max_env_steps: 396000
-  rollout_batch_size: 5
-  rollout_steps: 660
-  eval_batch_size: 10
+   # runner args
+   max_env_steps: 660000
+   rollout_batch_size: 5
+   rollout_steps: 660
+   eval_batch_size: 10
 
-  # misc
-  log_interval: 6600
-  save_interval: 660000
-  num_checkpoints: 0
-  eval_interval: 6600
-  eval_save_best: True
-  tensorboard: False
+   # misc
+   log_interval: 13200
+   save_interval: 660000
+   num_checkpoints: 0
+   eval_interval: 13200
+   eval_save_best: True
+   tensorboard: False
 
 #algo_config:
 #  # model args
-#  hidden_dim: 128
-#  activation: "relu"
+#  hidden_dim: 8
+#  activation: "tanh"
 #
 #  # loss args
 #  gamma: 0.98
-#  use_gae: True   # or False
-#  gae_lambda: 0.92
-#  use_clipped_value: False  # or True
-#  clip_param: 0.1
-#  target_kl: 1.0e-5
-#  entropy_coef: 0.003
+#  use_gae: True
+#  gae_lambda: 0.9
+#  clip_param: 0.2
+#  target_kl: 2.32e-2
+#  entropy_coef: 0.09
 #
 #  # optim args
-#  opt_epochs: 25
+#  opt_epochs: 20
 #  mini_batch_size: 256
-#  actor_lr: 7.2e-5
-#  critic_lr: 0.0266
+#  actor_lr: 0.0012
+#  critic_lr: 0.0012
 #
 #  # runner args
-#  max_env_steps: 216000
+#  max_env_steps: 660000
 #  rollout_batch_size: 5
 #  rollout_steps: 660
 #  eval_batch_size: 10
 #
 #  # misc
-#  log_interval: 6600
+#  log_interval: 13200
 #  save_interval: 660000
 #  num_checkpoints: 0
-#  eval_interval: 6600
+#  eval_interval: 13200
 #  eval_save_best: True
 #  tensorboard: False

examples/rl/config_overrides/quadrotor_2D_attitude/quadrotor_2D_attitude_track.yaml

@@ -61,8 +61,10 @@ task_config:
   obs_goal_horizon: 1
 
   # RL Reward
-  rew_state_weight: [10., .0, 10., .0, .0, 0.0]
-  rew_act_weight: [.0, .0]
+  rew_state_weight: [10., .1, 10., .1, .1, 0.001]
+  rew_act_weight: [.1, .1]
+#  rew_state_weight: [ 8.45, .48, 9.18, .81, .49, 0.001]
+#  rew_act_weight: [ .855, 1.194]
   rew_exponential: True
 
   disturbances: