Adds Hydra configuration preset overriding features.

docs/source/features/hydra.rst

@@ -127,3 +127,162 @@ the post init update is as follows:
 
 Here, when modifying ``env.decimation`` or ``env.sim.dt``, the user needs to give the updated ``env.sim.render_interval``,
 ``env.scene.height_scanner.update_period``, and ``env.scene.contact_forces.update_period`` as input as well.
+
+
+Group Override
+--------------
+Group override lets you swap out entire groups of environment- or agent-level settings in one go.
+Instead of overriding individual fields, you select a named preset defined under a ``variants`` mapping
+directly inside your config classes.
+
+
+Defining Variants
+^^^^^^^^^^^^^^^^^
+Declare alternatives under ``self.variants`` in your environment and agent configs. Each top-level key under
+``variants`` becomes a Hydra group (``env.<key>`` or ``agent.<key>``), and each nested key is a selectable option.
+
+Environment variants example:
+
+.. code-block:: python
+
+    @configclass
+    class ReachEnvCfg(ManagerBasedRLEnvCfg):
+        def __post_init__(self):
+            super().__post_init__()
+            # Share across all derived envs
+            self.variants = {
+                "observations": {
+                    "noise_less": NoiselessObservationsCfg(),
+                }
+            }
+
+    @configclass
+    class FrankaReachEnvCfg(ReachEnvCfg):
+        def __post_init__(self):
+            super().__post_init__()
+            variants = {
+                "actions.arm_action": {
+                    "joint_position_to_limit": mdp.JointPositionToLimitsActionCfg(
+                        asset_name="robot", joint_names=["panda_joint.*"]
+                    ),
+                    "relative_joint_position": mdp.RelativeJointPositionActionCfg(
+                        asset_name="robot", joint_names=["panda_joint.*"], scale=0.2
+                    ),
+                }
+            }
+            self.variants.update(variants)
+
+RSL-RL agent variants example:
+
+.. code-block:: python
+
+    @configclass
+    class FrankaReachPPORunnerCfg(RslRlOnPolicyRunnerCfg):
+        num_steps_per_env = 24
+        ...
+        policy = RslRlPpoActorCriticCfg(
+            ...
+        )
+        algorithm = RslRlPpoAlgorithmCfg(
+            ...
+        )
+        variants = {
+            "policy": {
+                "large_network": RslRlPpoActorCriticCfg(
+                    actor_hidden_dims=[512, 256, 128, 64], critic_hidden_dims=[512, 256, 128, 64], ...
+                ),
+                "medium_network": RslRlPpoActorCriticCfg(
+                    actor_hidden_dims=[256, 128, 64], critic_hidden_dims=[256, 128, 64], ...
+                ),
+            },
+            "algorithm": {
+                "small_batch_lr": RslRlPpoAlgorithmCfg(num_mini_batches=16, learning_rate=1.0e-4, ...),
+            },
+        }
+
+
+RL Games agent variants example:
+
+.. code-block:: yaml
+
+   params:
+     env: ...
+     config: ...
+     network:
+       ...
+       mlp:
+         units: [64, 64]
+         activation: elu
+         d2rl: False
+
+   variants:
+     params.network.mlp:
+       large_network:
+         units: [256, 128, 64]
+         activation: elu
+         d2rl: False
+
+The above defines a selectable group at ``agent.params.network.mlp`` with option ``large_network``.
+
+
+
+
+
+Override Syntax
+^^^^^^^^^^^^^^^
+Select one preset per group via Hydra-style CLI flags.
+
+.. tab-set::
+   :sync-group: rl-override
+
+   .. tab-item:: rsl_rl
+      :sync: rsl_rl
+
+      .. code-block:: bash
+
+         python scripts/reinforcement_learning/rsl_rl/train.py \
+           --task=Isaac-Reach-Franka-v0 \
+           --headless \
+           env.observations=noise_less \
+           env.actions.arm_action=relative_joint_position \
+           agent.policy=large_network
+
+      Hydra replaces:
+
+      .. list-table::
+         :widths: 30 70
+         :header-rows: 1
+
+         * - CLI key
+           - Resolved variant node
+         * - ``env.observations``
+           - ``ReachEnvCfg.variants["observations"]["noise_less"]``
+         * - ``env.actions.arm_action``
+           - ``FrankaReachEnvCfg.variants["actions.arm_action"]["relative_joint_position"]``
+         * - ``agent.policy``
+           - ``FrankaReachPPORunnerCfg.variants["policy"]["large_network"]``
+
+   .. tab-item:: rl_games
+      :sync: rl_games
+
+      .. code-block:: bash
+
+         python scripts/reinforcement_learning/rl_games/train.py \
+           --task=Isaac-Reach-Franka-v0 \
+           --headless \
+           env.observations=noise_less \
+           env.actions.arm_action=relative_joint_position \
+           agent.params.network.mlp=large_network
+
+      Hydra replaces:
+
+      .. list-table::
+         :widths: 35 65
+         :header-rows: 1
+
+         * - CLI key
+           - Resolved variant node
+         * - ``agent.params.network.mlp``
+           - ``variants["params.network.mlp"]["large_network"]`` (from RL Games YAML)
+
+These flags let you switch qualitative modes of your experiments with a single option per group.

source/isaaclab_tasks/config/extension.toml

@@ -1,7 +1,7 @@
 [package]
 
 # Note: Semantic Versioning is used: https://semver.org/
-version = "0.11.1"
+version = "0.12.0"
 
 # Description
 title = "Isaac Lab Environments"

source/isaaclab_tasks/docs/CHANGELOG.rst

@@ -1,6 +1,15 @@
 Changelog
 ---------
 
+0.12.0 (2025-10-15)
+~~~~~~~~~~~~~~~~~~~~
+
+Changed
+^^^^^^^
+
+* Add new feature that support hydra group config override, and provide example at Isaac-Reach-Franka-v0 env
+
+
 0.11.1 (2025-09-24)
 ~~~~~~~~~~~~~~~~~~~~
 

source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/reach/config/franka/agents/rl_games_ppo_cfg.yaml

@@ -81,3 +81,15 @@ params:
     clip_value: True
     clip_actions: False
     bounds_loss_coef: 0.0001
+
+variants:
+  params.network.mlp:
+    large_network:
+      units: [256, 128, 64]
+      activation: elu
+      d2rl: False
+
+      initializer:
+        name: default
+      regularizer:
+        name: None

source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/reach/config/franka/agents/rsl_rl_ppo_cfg.py

@@ -37,3 +37,55 @@ class FrankaReachPPORunnerCfg(RslRlOnPolicyRunnerCfg):
         desired_kl=0.01,
         max_grad_norm=1.0,
     )
+    variants = {
+        "policy": {
+            "large_network": RslRlPpoActorCriticCfg(
+                init_noise_std=1.0,
+                actor_hidden_dims=[512, 256, 128, 64],
+                critic_hidden_dims=[512, 256, 128, 64],
+                activation="elu",
+            ),
+            "medium_network": RslRlPpoActorCriticCfg(
+                init_noise_std=1.0,
+                actor_hidden_dims=[256, 128, 64],
+                critic_hidden_dims=[256, 128, 64],
+                activation="elu",
+            ),
+            "small_network": RslRlPpoActorCriticCfg(
+                init_noise_std=1.0,
+                actor_hidden_dims=[128, 64],
+                critic_hidden_dims=[128, 64],
+                activation="elu",
+            ),
+        },
+        "algorithm": {
+            "large_batch_lr": RslRlPpoAlgorithmCfg(
+                value_loss_coef=1.0,
+                use_clipped_value_loss=True,
+                clip_param=0.2,
+                entropy_coef=0.001,
+                num_learning_epochs=8,
+                num_mini_batches=2,
+                learning_rate=1.0e-3,
+                schedule="adaptive",
+                gamma=0.99,
+                lam=0.95,
+                desired_kl=0.01,
+                max_grad_norm=1.0,
+            ),
+            "small_batch_lr": RslRlPpoAlgorithmCfg(
+                value_loss_coef=1.0,
+                use_clipped_value_loss=True,
+                clip_param=0.2,
+                entropy_coef=0.001,
+                num_learning_epochs=8,
+                num_mini_batches=16,
+                learning_rate=1.0e-4,
+                schedule="adaptive",
+                gamma=0.99,
+                lam=0.95,
+                desired_kl=0.01,
+                max_grad_norm=1.0,
+            ),
+        },
+    }

source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/reach/config/franka/joint_pos_env_cfg.py

@@ -43,6 +43,19 @@ def __post_init__(self):
         self.commands.ee_pose.body_name = "panda_hand"
         self.commands.ee_pose.ranges.pitch = (math.pi, math.pi)
 
+        variants = {
+            "actions.arm_action": {
+                "joint_position_to_limit": mdp.JointPositionToLimitsActionCfg(
+                    asset_name="robot", joint_names=["panda_joint.*"]
+                ),
+                "relative_joint_position": mdp.RelativeJointPositionActionCfg(
+                    asset_name="robot", joint_names=["panda_joint.*"], scale=0.2
+                ),
+            }
+        }
+
+        self.variants.update(variants)
+
 
 @configclass
 class FrankaReachEnvCfg_PLAY(FrankaReachEnvCfg):

source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/reach/reach_env_cfg.py

@@ -116,6 +116,28 @@ def __post_init__(self):
     policy: PolicyCfg = PolicyCfg()
 
 
+@configclass
+class NoiselessObservationsCfg:
+    """Noise less observation specifications for the MDP."""
+
+    @configclass
+    class PolicyCfg(ObsGroup):
+        """Observations for policy group."""
+
+        # observation terms (order preserved)
+        joint_pos = ObsTerm(func=mdp.joint_pos_rel)
+        joint_vel = ObsTerm(func=mdp.joint_vel_rel)
+        pose_command = ObsTerm(func=mdp.generated_commands, params={"command_name": "ee_pose"})
+        actions = ObsTerm(func=mdp.last_action)
+
+        def __post_init__(self):
+            self.enable_corruption = False
+            self.concatenate_terms = True
+
+    # observation groups
+    policy: PolicyCfg = PolicyCfg()
+
+
 @configclass
 class EventCfg:
     """Configuration for events."""
@@ -227,3 +249,5 @@ def __post_init__(self):
                 ),
             },
         )
+        # variants defined at base env will be shared across all derived robot-specific envs
+        self.variants = {"observations": {"noise_less": NoiselessObservationsCfg()}}