Share the manager section declarations through term factories

The reorient manager cfgs repeated the same command, reward, and
termination declarations per robot with one or two differing scalars.
Move the defaults into two term factories and the invariant termination
section in reorient_task_base; each variant now declares a small plain
section class stating only its deltas (Allegro tolerance 0.2, OpenAI
tolerance 0.4 with fall penalty and noisy-EMA action source). The
camera manager cfg inherits the sections instead of re-declaring them.

Author: hujc7

source/isaaclab_tasks/isaaclab_tasks/core/reorient/config/allegro_hand/allegro_hand_manager_env_cfg.py

@@ -11,9 +11,7 @@
 from isaaclab.managers import EventTermCfg as EventTerm
 from isaaclab.managers import ObservationGroupCfg as ObsGroup
 from isaaclab.managers import ObservationTermCfg as ObsTerm
-from isaaclab.managers import RewardTermCfg as RewTerm
 from isaaclab.managers import SceneEntityCfg
-from isaaclab.managers import TerminationTermCfg as DoneTerm
 from isaaclab.scene import InteractiveSceneCfg
 from isaaclab.utils.configclass import configclass
 
@@ -28,8 +26,9 @@
 from isaaclab_tasks.core.reorient.reorient_task_base import (
     ALLEGRO_ACTUATED_JOINT_NAMES,
     ALLEGRO_FINGERTIP_BODY_NAMES,
-    GOAL_MARKER_POSITION,
-    IN_HAND_POS_OFFSET,
+    ReorientTerminationsCfg,
+    reorient_goal_command,
+    reorient_reward_term,
 )
 from isaaclab_tasks.utils import PresetCfg
 
@@ -69,18 +68,9 @@ def set_num_envs(self, num_envs: int) -> None:
 
 @configclass
 class CommandsCfg:
-    """Object pose goal matching the Direct in-hand target."""
-
-    object_pose = mdp.ReorientEpisodeCommandCfg(
-        asset_name="object",
-        init_pos_offset=IN_HAND_POS_OFFSET,
-        update_goal_on_success=True,
-        orientation_success_threshold=0.2,
-        make_quat_unique=False,
-        fixed_marker_pos=GOAL_MARKER_POSITION,
-        goal_pose_visualizer_cfg=GOAL_OBJECT_CFG,
-        debug_vis=True,
-    )
+    """The reorient goal command with the Allegro marker and tolerance."""
+
+    object_pose = reorient_goal_command(orientation_success_threshold=0.2, goal_pose_visualizer_cfg=GOAL_OBJECT_CFG)
 
 
 @configclass
@@ -231,41 +221,9 @@ class EventCfg:
 
 @configclass
 class RewardsCfg:
-    """Direct-compatible reward and success accounting."""
-
-    reorient = RewTerm(
-        func=mdp.ReorientReward,
-        weight=1.0,
-        params={
-            "command_name": "object_pose",
-            "distance_scale": -10.0,
-            "rotation_scale": 1.0,
-            "rotation_epsilon": 0.1,
-            "action_penalty_scale": -0.0002,
-            "success_tolerance": 0.2,
-            "success_bonus": 250.0,
-            "fall_distance": 0.24,
-            "fall_penalty": 0.0,
-            "averaging_factor": 0.1,
-            "success_count_threshold": 1,
-            "object_cfg": SceneEntityCfg("object"),
-        },
-    )
+    """The reorient reward with the Allegro success tolerance."""
 
-
-@configclass
-class TerminationsCfg:
-    """Termination conditions matching the Direct task."""
-
-    object_out_of_reach = DoneTerm(
-        func=mdp.object_reorientation_out_of_reach,
-        params={
-            "threshold": 0.24,
-            "command_name": "object_pose",
-            "object_cfg": SceneEntityCfg("object"),
-        },
-    )
-    time_out = DoneTerm(func=mdp.time_out, time_out=True)
+    reorient = reorient_reward_term(success_tolerance=0.2)
 
 
 @configclass
@@ -277,7 +235,8 @@ class AllegroCubeEnvCfg(AllegroHandTaskCfgBase, ManagerBasedRLEnvCfg):
     actions: ActionsCfg = ActionsCfg()
     commands: CommandsCfg = CommandsCfg()
     rewards: RewardsCfg = RewardsCfg()
-    terminations: TerminationsCfg = TerminationsCfg()
+    # the shared termination section from isaaclab_tasks.core.reorient.reorient_task_base
+    terminations: ReorientTerminationsCfg = ReorientTerminationsCfg()
     events: EventCfg = EventCfg()
 
     enable_domain_randomization: bool = False

source/isaaclab_tasks/isaaclab_tasks/core/reorient/config/shadow_hand/shadow_hand_camera_manager_env_cfg.py

@@ -18,12 +18,9 @@
 )
 from isaaclab_tasks.core.reorient.config.shadow_hand.shadow_hand_manager_env_cfg import (
     ActionsCfg,
-    CommandsCfg,
     EventCfg,
     FullStateWithoutActionCfg,
-    RewardsCfg,
     ShadowHandManagerEnvCfg,
-    TerminationsCfg,
     _ShadowHandManagerSceneCfg,
 )
 from isaaclab_tasks.core.reorient.reorient_task_base import (
@@ -122,9 +119,8 @@ class ShadowHandCameraManagerEnvCfg(ShadowHandManagerEnvCfg):
     scene: ShadowHandCameraManagerSceneCfg = ShadowHandCameraManagerSceneCfg()
     observations: CameraObservationsCfg = CameraObservationsCfg()
     actions: ActionsCfg = ActionsCfg()
-    commands: CommandsCfg = CommandsCfg()
-    rewards: RewardsCfg = RewardsCfg()
-    terminations: TerminationsCfg = TerminationsCfg()
+    # commands/rewards/terminations are inherited from ShadowHandManagerEnvCfg (the
+    # shared sections from isaaclab_tasks.core.reorient.reorient_task_base)
     events: EventCfg = EventCfg()
     feature_extractor: FeatureExtractorCfg = FeatureExtractorCfg()
 

source/isaaclab_tasks/isaaclab_tasks/core/reorient/config/shadow_hand/shadow_hand_manager_env_cfg.py

@@ -11,7 +11,6 @@
 from isaaclab.managers import EventTermCfg as EventTerm
 from isaaclab.managers import ObservationGroupCfg as ObsGroup
 from isaaclab.managers import ObservationTermCfg as ObsTerm
-from isaaclab.managers import RewardTermCfg as RewTerm
 from isaaclab.managers import SceneEntityCfg
 from isaaclab.managers import TerminationTermCfg as DoneTerm
 from isaaclab.scene import InteractiveSceneCfg
@@ -32,10 +31,11 @@
     ShadowHandTaskCfgBase,
 )
 from isaaclab_tasks.core.reorient.reorient_task_base import (
-    GOAL_MARKER_POSITION,
-    IN_HAND_POS_OFFSET,
     SHADOW_ACTUATED_JOINT_NAMES,
     SHADOW_FINGERTIP_BODY_NAMES,
+    ReorientTerminationsCfg,
+    reorient_goal_command,
+    reorient_reward_term,
 )
 from isaaclab_tasks.utils import PresetCfg
 
@@ -68,22 +68,6 @@ class ShadowHandManagerSceneCfg(PresetCfg):
     default = physx
 
 
-@configclass
-class CommandsCfg:
-    """Object pose goal matching the Direct in-hand target."""
-
-    object_pose = mdp.ReorientEpisodeCommandCfg(
-        asset_name="object",
-        init_pos_offset=IN_HAND_POS_OFFSET,
-        update_goal_on_success=True,
-        orientation_success_threshold=0.1,
-        make_quat_unique=False,
-        fixed_marker_pos=GOAL_MARKER_POSITION,
-        goal_pose_visualizer_cfg=GOAL_OBJECT_CFG,
-        debug_vis=True,
-    )
-
-
 @configclass
 class ActionsCfg:
     """Twenty actuated Shadow Hand joints."""
@@ -171,42 +155,17 @@ class EventCfg:
 
 
 @configclass
-class RewardsCfg:
-    """Direct-compatible reward and success accounting."""
+class CommandsCfg:
+    """The reorient goal command with the Shadow Hand goal marker."""
 
-    reorient = RewTerm(
-        func=mdp.ReorientReward,
-        weight=1.0,
-        params={
-            "command_name": "object_pose",
-            "distance_scale": -10.0,
-            "rotation_scale": 1.0,
-            "rotation_epsilon": 0.1,
-            "action_penalty_scale": -0.0002,
-            "success_tolerance": 0.1,
-            "success_bonus": 250.0,
-            "fall_distance": 0.24,
-            "fall_penalty": 0.0,
-            "averaging_factor": 0.1,
-            "success_count_threshold": 1,
-            "object_cfg": SceneEntityCfg("object"),
-        },
-    )
+    object_pose = reorient_goal_command(goal_pose_visualizer_cfg=GOAL_OBJECT_CFG)
 
 
 @configclass
-class TerminationsCfg:
-    """Termination conditions matching the Direct task."""
+class RewardsCfg:
+    """Direct-compatible reward and success accounting (the factory defaults)."""
 
-    object_out_of_reach = DoneTerm(
-        func=mdp.object_reorientation_out_of_reach,
-        params={
-            "threshold": 0.24,
-            "command_name": "object_pose",
-            "object_cfg": SceneEntityCfg("object"),
-        },
-    )
-    time_out = DoneTerm(func=mdp.time_out, time_out=True)
+    reorient = reorient_reward_term()
 
 
 @configclass
@@ -218,7 +177,8 @@ class ShadowHandManagerEnvCfg(ShadowHandTaskCfgBase, ManagerBasedRLEnvCfg):
     actions: ActionsCfg = ActionsCfg()
     commands: CommandsCfg = CommandsCfg()
     rewards: RewardsCfg = RewardsCfg()
-    terminations: TerminationsCfg = TerminationsCfg()
+    # the shared termination section from isaaclab_tasks.core.reorient.reorient_task_base
+    terminations: ReorientTerminationsCfg = ReorientTerminationsCfg()
     events: EventCfg = EventCfg()
 
     def __post_init__(self):
@@ -228,22 +188,6 @@ def __post_init__(self):
         self.viewer.eye = (2.0, 2.0, 2.0)
 
 
-@configclass
-class OpenAICommandsCfg:
-    """OpenAI goal command with its wider success tolerance."""
-
-    object_pose = mdp.ReorientEpisodeCommandCfg(
-        asset_name="object",
-        init_pos_offset=IN_HAND_POS_OFFSET,
-        update_goal_on_success=True,
-        orientation_success_threshold=0.4,
-        make_quat_unique=False,
-        fixed_marker_pos=GOAL_MARKER_POSITION,
-        goal_pose_visualizer_cfg=GOAL_OBJECT_CFG,
-        debug_vis=True,
-    )
-
-
 @configclass
 class OpenAIActionsCfg:
     """OpenAI actions with Direct-compatible EMA and stateful noise."""
@@ -345,29 +289,18 @@ class OpenAIEventCfg(PresetCfg):
     default = physx
 
 
+@configclass
+class OpenAICommandsCfg:
+    """OpenAI goal command with its wider success tolerance."""
+
+    object_pose = reorient_goal_command(orientation_success_threshold=0.4, goal_pose_visualizer_cfg=GOAL_OBJECT_CFG)
+
+
 @configclass
 class OpenAIRewardsCfg:
-    """Direct-compatible OpenAI reward and success accounting."""
+    """OpenAI reward: wider tolerance, fall penalty, and noisy-EMA action source."""
 
-    reorient = RewTerm(
-        func=mdp.ReorientReward,
-        weight=1.0,
-        params={
-            "command_name": "object_pose",
-            "distance_scale": -10.0,
-            "rotation_scale": 1.0,
-            "rotation_epsilon": 0.1,
-            "action_penalty_scale": -0.0002,
-            "success_tolerance": 0.4,
-            "success_bonus": 250.0,
-            "fall_distance": 0.24,
-            "fall_penalty": -50.0,
-            "averaging_factor": 0.1,
-            "success_count_threshold": 1,
-            "action_name": "joint_pos",
-            "object_cfg": SceneEntityCfg("object"),
-        },
-    )
+    reorient = reorient_reward_term(success_tolerance=0.4, fall_penalty=-50.0, action_name="joint_pos")
 
 
 @configclass

source/isaaclab_tasks/isaaclab_tasks/core/reorient/reorient_task_base.py

@@ -6,11 +6,19 @@
 """Structural definitions shared by the reorientation task family.
 
 Joint/body name lists, marker geometry, and reset-pose offsets consumed by both
-the Direct and manager-based variants. Scalar task parameters are defined
-per-paradigm in the respective environment configurations, following the
-convention of the other core tasks.
+the Direct and manager-based variants, plus the manager term factories and the
+shared termination section used by the robot-specific manager tasks. The
+factory defaults carry the Shadow Hand state-task values; each variant declares
+its section class stating only its deltas.
 """
 
+from isaaclab.managers import RewardTermCfg as RewTerm
+from isaaclab.managers import SceneEntityCfg
+from isaaclab.managers import TerminationTermCfg as DoneTerm
+from isaaclab.utils.configclass import configclass
+
+import isaaclab_tasks.core.reorient.mdp as mdp
+
 GOAL_MARKER_POSITION: tuple[float, float, float] = (-0.2, -0.45, 0.68)
 """Fixed goal-marker display position [m], environment frame (state-based tasks)."""
 
@@ -92,3 +100,61 @@
     "robot0_THJ0",
 ]
 """Shadow Hand actuated joint names, in the Direct task's actuation order."""
+
+
+def reorient_goal_command(**overrides) -> "mdp.ReorientEpisodeCommandCfg":
+    """Build the object-pose goal command with the Direct-parity defaults.
+
+    Keyword overrides replace the corresponding command fields; robot-specific
+    variants declare the goal marker (``goal_pose_visualizer_cfg``) and their
+    success threshold this way at the declaration site.
+    """
+    return mdp.ReorientEpisodeCommandCfg(
+        asset_name="object",
+        init_pos_offset=IN_HAND_POS_OFFSET,
+        update_goal_on_success=True,
+        orientation_success_threshold=0.1,
+        make_quat_unique=False,
+        fixed_marker_pos=GOAL_MARKER_POSITION,
+        debug_vis=True,
+    ).replace(**overrides)
+
+
+def reorient_reward_term(**param_overrides) -> RewTerm:
+    """Build the :class:`~isaaclab_tasks.core.reorient.mdp.ReorientReward` term.
+
+    The parameter defaults are the Direct-parity values of the Shadow Hand
+    state task; keyword overrides are merged on top, so variants state only
+    their deltas at the declaration site.
+    """
+    params = {
+        "command_name": "object_pose",
+        "distance_scale": -10.0,
+        "rotation_scale": 1.0,
+        "rotation_epsilon": 0.1,
+        "action_penalty_scale": -0.0002,
+        "success_tolerance": 0.1,
+        "success_bonus": 250.0,
+        "fall_distance": 0.24,
+        "fall_penalty": 0.0,
+        "averaging_factor": 0.1,
+        "success_count_threshold": 1,
+        "object_cfg": SceneEntityCfg("object"),
+    }
+    params.update(param_overrides)
+    return RewTerm(func=mdp.ReorientReward, weight=1.0, params=params)
+
+
+@configclass
+class ReorientTerminationsCfg:
+    """Termination conditions matching the Direct task."""
+
+    object_out_of_reach = DoneTerm(
+        func=mdp.object_reorientation_out_of_reach,
+        params={
+            "threshold": 0.24,
+            "command_name": "object_pose",
+            "object_cfg": SceneEntityCfg("object"),
+        },
+    )
+    time_out = DoneTerm(func=mdp.time_out, time_out=True)