Implemented the rewards hyperparameters

amp_rsl_rl/runners/amp_on_policy_runner.py

@@ -354,8 +354,8 @@ def update_run_name_with_sequence(prefix: str) -> None:
                     mean_task_reward_log += rewards.mean().item()
                     mean_style_reward_log += style_rewards.mean().item()
 
-                    # Combine the task and style rewards (TODO this can be a hyperparameters)
-                    rewards = 0.5 * rewards + 0.5 * style_rewards
+                    # Combine the task and style rewards
+                    rewards = self.cfg['task_reward_weight'] * rewards + self.cfg['style_reward_weight'] * style_rewards
 
                     self.alg.process_env_step(rewards, dones, infos)
                     self.alg.process_amp_step(next_amp_obs)

example/agents_config.py

@@ -0,0 +1,26 @@
+from isaaclab.utils import configclass
+from isaaclab_rl.rsl_rl import RslRlOnPolicyRunnerCfg
+
+@configclass
+class YourConfigName(RslRlOnPolicyRunnerCfg):
+    experiment_name = "XXX"
+    discriminator = {
+        "hidden_dims": [512, 256],
+        "reward_scale": 1.0,
+        "loss_type": "BCEWithLogits"  # Choose between BCEWithLogits or Wasserstein
+    }
+
+    # Weights for combining task and style rewards
+    task_reward_weight = 0.5
+    style_reward_weight = 0.5
+
+    amp_data_path = "path of the dataset folder"
+    dataset_names = ["files", 
+                     "without", 
+                     ".npy",
+                     ]
+    dataset_weights = [1.0 for i in range(len(dataset_names))]
+    slow_down_factor = 1.0
+
+    def __post_init__(self):
+        self.algorithm.class_name = "AMP_PPO"