feat: adds a configuration example for Student-Teacher Distillation

Author: fan-ziqi

README.md

@@ -240,10 +240,24 @@ Others (**Experimental**)
   ```bash
   # Train
   python scripts/reinforcement_learning/rsl_rl/train.py --task=RobotLab-Isaac-Velocity-Rough-Anymal-D-v0 --headless --agent=rsl_rl_with_symmetry_cfg_entry_point --run_name=ppo_with_symmetry_data_augmentation agent.algorithm.symmetry_cfg.use_data_augmentation=true
-  # Train
+
+  # Play
   python scripts/reinforcement_learning/rsl_rl/play.py --task=RobotLab-Isaac-Velocity-Rough-Anymal-D-v0 --agent=rsl_rl_with_symmetry_cfg_entry_point --run_name=ppo_with_symmetry_data_augmentation agent.algorithm.symmetry_cfg.use_data_augmentation=true
   ```
 
+- Training and distilling Anymal D
+
+  ```bash
+  # Train the teacher agent
+  python scripts/reinforcement_learning/rsl_rl/train.py --task=RobotLab-Isaac-Velocity-Flat-Anymal-D-v0 --headless
+
+  # Distill the teacher agent into a student agent
+  python scripts/reinforcement_learning/rsl_rl/train.py --task=RobotLab-Isaac-Velocity-Flat-Anymal-D-v0 --headless --agent=rsl_rl_distillation_cfg_entry_point --load_run teacher_run_folder_name
+
+  # Play the student agent
+  python scripts/reinforcement_learning/rsl_rl/play.py --task=RobotLab-Isaac-Velocity-Flat-Anymal-D-v0 --num_envs 64 --agent rsl_rl_distillation_cfg_entry_point
+  ```
+
 > [!NOTE]
 > If you want to control a **SINGLE ROBOT** with the keyboard during playback, add `--keyboard` at the end of the play script.
 >

source/robot_lab/robot_lab/tasks/manager_based/locomotion/velocity/config/quadruped/anymal_d/__init__.py

@@ -18,6 +18,9 @@
         "rsl_rl_cfg_entry_point": f"{agents.__name__}.rsl_rl_ppo_cfg:AnymalDFlatPPORunnerCfg",
         "rsl_rl_with_symmetry_cfg_entry_point": f"{agents.__name__}.rsl_rl_ppo_cfg:AnymalDFlatPPORunnerWithSymmetryCfg",
         "cusrl_cfg_entry_point": f"{agents.__name__}.cusrl_ppo_cfg:AnymalDFlatTrainerCfg",
+        "rsl_rl_distillation_cfg_entry_point": (
+            f"{agents.__name__}.rsl_rl_distillation_cfg:AnymalDFlatDistillationRunnerCfg"
+        ),
         "cusrl_with_symmetry_cfg_entry_point": (
             f"{agents.__name__}.cusrl_ppo_cfg:AnymalDFlatTrainerCfgWithSymmetryAugmentation"
         ),

source/robot_lab/robot_lab/tasks/manager_based/locomotion/velocity/config/quadruped/anymal_d/agents/rsl_rl_distillation_cfg.py

@@ -0,0 +1,35 @@
+# Copyright (c) 2022-2025, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
+# All rights reserved.
+#
+# SPDX-License-Identifier: BSD-3-Clause
+
+from isaaclab.utils import configclass
+
+from isaaclab_rl.rsl_rl import (
+    RslRlDistillationAlgorithmCfg,
+    RslRlDistillationRunnerCfg,
+    RslRlDistillationStudentTeacherCfg,
+)
+
+
+@configclass
+class AnymalDFlatDistillationRunnerCfg(RslRlDistillationRunnerCfg):
+    num_steps_per_env = 120
+    max_iterations = 300
+    save_interval = 50
+    experiment_name = "anymal_d_flat"
+    obs_groups = {"policy": ["policy"], "teacher": ["policy"]}
+    policy = RslRlDistillationStudentTeacherCfg(
+        init_noise_std=0.1,
+        noise_std_type="scalar",
+        student_obs_normalization=False,
+        teacher_obs_normalization=False,
+        student_hidden_dims=[512, 256, 128],
+        teacher_hidden_dims=[512, 256, 128],
+        activation="elu",
+    )
+    algorithm = RslRlDistillationAlgorithmCfg(
+        num_learning_epochs=2,
+        learning_rate=1.0e-3,
+        gradient_length=15,
+    )

source/robot_lab/robot_lab/tasks/manager_based/locomotion/velocity/config/quadruped/anymal_d/rough_env_cfg.py

@@ -30,8 +30,8 @@ def __post_init__(self):
         self.observations.policy.base_ang_vel.scale = 0.25
         self.observations.policy.joint_pos.scale = 1.0
         self.observations.policy.joint_vel.scale = 0.05
-        self.observations.policy.base_lin_vel = None
-        self.observations.policy.height_scan = None
+        # self.observations.policy.base_lin_vel = None
+        # self.observations.policy.height_scan = None
 
         # ------------------------------Actions------------------------------
         # reduce action scale