Fixes configurations for Stable-Baselines3 workflow (#61)

* fixes sb3 configs for missing agent seed
* fixes encoding issue when loading yaml configs
* adds launch.json task for running environment

.vscode/launch.json

@@ -24,5 +24,19 @@
             "port": 3000,
             "host": "localhost"
         },
+        {
+            "name": "Python: Run Environment",
+            "type": "python",
+            "request": "launch",
+            "args" : ["--task", "Isaac-Reach-Franka-v0"],
+            "program": "${file}",
+            "console": "integratedTerminal",
+            "env": {
+                "EXP_PATH": "${workspaceFolder}/_isaac_sim/apps",
+                "RESOURCE_NAME": "IsaacSim"
+            },
+            "envFile": "${workspaceFolder}/.vscode/.python.env",
+            "preLaunchTask": "setup_python_env"
+        },
     ]
 }

source/extensions/omni.isaac.orbit/omni/isaac/orbit/utils/math.py

@@ -329,7 +329,7 @@ def axis_angle_from_quat(quat: torch.Tensor, eps: float = 1.0e-6) -> torch.Tenso
     # Thus, axis-angle is [q_x, q_y, q_z] / (sin(theta/2) / theta)
     # When theta = 0, (sin(theta/2) / theta) is undefined
     # However, as theta --> 0, we can use the Taylor approximation 1/2 - theta^2 / 48
-    quat = quat * (1.0 - 2. * (quat[..., 0:1] < 0.0))
+    quat = quat * (1.0 - 2.0 * (quat[..., 0:1] < 0.0))
     mag = torch.linalg.norm(quat[..., 1:], dim=1)
     half_angle = torch.atan2(mag, quat[..., 0])
     angle = 2.0 * half_angle

source/extensions/omni.isaac.orbit_envs/data/sb3/ant_ppo.yaml

 # Reference: https://github.com/DLR-RM/rl-baselines3-zoo/blob/master/hyperparams/ppo.yml#L161
+seed: 42
 
 n_timesteps: !!float 2e6
 policy: 'MlpPolicy'

source/extensions/omni.isaac.orbit_envs/data/sb3/cabinet_ppo.yaml

 # Reference: https://github.com/DLR-RM/rl-baselines3-zoo/blob/master/hyperparams/ppo.yml#L32
-# 512×1500×64
 seed: 42
-n_timesteps: !!float 49152000
+
+# 512×1500×64
+n_timesteps: 49152000
 policy: 'MlpPolicy'
 n_steps: 64
 batch_size: 4096

source/extensions/omni.isaac.orbit_envs/data/sb3/cartpole_sac.yaml

 # Reference: https://github.com/DLR-RM/rl-baselines3-zoo/blob/master/hyperparams/ppo.yml#L32
-seed: 67854
+seed: 42
 
 # 512×500×16
 n_timesteps: 4096000

source/extensions/omni.isaac.orbit_envs/data/sb3/humanoid_ppo.yaml

 # Reference: https://github.com/DLR-RM/rl-baselines3-zoo/blob/master/hyperparams/ppo.yml#L457
+seed: 42
 
 policy: 'MlpPolicy'
 n_timesteps: !!float 1e7

source/extensions/omni.isaac.orbit_envs/data/sb3/reach_ppo.yaml

 # Reference: https://github.com/DLR-RM/rl-baselines3-zoo/blob/master/hyperparams/ppo.yml#L32
+seed: 42
+
 # epoch * n_steps * nenvs: 500×512*8*8
-n_timesteps: !!float 16384000
+n_timesteps: 16384000
 policy: 'MlpPolicy'
 n_steps: 64
 # mini batch size: num_envs * nsteps / nminibatches 2048×512÷2048
@@ -16,13 +18,9 @@ policy_kwargs: "dict(
                   activation_fn=nn.ELU,
                   net_arch=[32, 32, dict(pi=[256, 128, 64], vf=[256, 128, 64])]
                 )"
-
-
 target_kl: 0.01
 max_grad_norm: 1.0
 
-
-
 # # Uses VecNormalize class to normalize obs
 # normalize_input: True
 # # Uses VecNormalize class to normalize rew

source/standalone/workflows/rl_games/config.py

@@ -41,7 +41,7 @@ def parse_rlg_cfg(task_name) -> dict:
         raise ValueError(f"Task not found: {task_name}")
 
     # parse agent configuration
-    with open(config_file) as f:
+    with open(config_file, encoding="utf-8") as f:
         cfg = yaml.load(f, Loader=yaml.Loader)
 
     return cfg

source/standalone/workflows/rsl_rl/config.py

@@ -42,7 +42,7 @@ def parse_rslrl_cfg(task_name) -> dict:
         raise ValueError(f"Task not found: {task_name}")
 
     # parse agent configuration
-    with open(config_file) as f:
+    with open(config_file, encoding="utf-8") as f:
         cfg = yaml.load(f, Loader=yaml.FullLoader)
 
     return cfg

source/standalone/workflows/sb3/config.py

@@ -40,7 +40,7 @@ def parse_sb3_cfg(task_name) -> dict:
         raise ValueError(f"Task not found: {task_name}. Configurations exist for {SB3_PPO_CONFIG_FILE.keys()}")
 
     # parse agent configuration
-    with open(config_file) as f:
+    with open(config_file, encoding="utf-8") as f:
         cfg = yaml.load(f, Loader=yaml.FullLoader)
     # check config is valid
     if cfg is None:

source/standalone/workflows/skrl/config.py

@@ -48,7 +48,7 @@ def parse_skrl_cfg(task_name) -> dict:
         raise ValueError(f"Task not found: {task_name}")
 
     # parse agent configuration
-    with open(config_file) as f:
+    with open(config_file, encoding="utf-8") as f:
         cfg = yaml.load(f, Loader=yaml.Loader)
 
     return cfg