Use the shared quaternion error metric and drop the direct_ prefixes

evaluate_reorient_success now computes the orientation error through
isaaclab.utils.math.quat_error_magnitude instead of a local formula:
the shared utility is torch.jit-scriptable (verified by compiling a
scripted caller from a source file) and decision-identical at the task
tolerances — zero success-threshold flips over 20k random pairs, with
differences up to 9e-4 rad only near pi where the shared atan2-based
form is the more accurate one. The local
direct_reorient_rotation_distance is deleted.

Rename direct_reorient_reward to reorient_reward: shared symbols carry
no paradigm prefix. The DirectReorientReward and DirectReorientTimeout
manager terms keep the prefix deliberately — there "Direct" names the
Direct-compatible contract they implement.

Author: hujc7

source/isaaclab_tasks/isaaclab_tasks/core/reorient/mdp/__init__.pyi

@@ -24,9 +24,8 @@ __all__ = [
     "success_bonus",
     "track_orientation_inv_l2",
     "track_pos_l2",
-    "direct_reorient_rotation_distance",
     "evaluate_reorient_success",
-    "direct_reorient_reward",
+    "reorient_reward",
     "DirectReorientReward",
     "max_consecutive_success",
     "object_away_from_goal",
@@ -55,8 +54,7 @@ from .observations import (
 )
 from .rewards import (
     DirectReorientReward,
-    direct_reorient_reward,
-    direct_reorient_rotation_distance,
+    reorient_reward,
     evaluate_reorient_success,
     success_bonus,
     track_orientation_inv_l2,

source/isaaclab_tasks/isaaclab_tasks/core/reorient/mdp/rewards.py

@@ -140,21 +140,6 @@ def track_orientation_inv_l2(
     return 1.0 / (dtheta + rot_eps)
 
 
-@torch.jit.script
-def direct_reorient_rotation_distance(object_quat: torch.Tensor, target_quat: torch.Tensor) -> torch.Tensor:
-    """Compute the Direct reorientation orientation distance [rad].
-
-    Args:
-        object_quat: Object ``(x, y, z, w)`` orientations.
-        target_quat: Target ``(x, y, z, w)`` orientations.
-
-    Returns:
-        Per-environment orientation distances [rad], in ``[0, pi]``.
-    """
-    quat_diff = math_utils.quat_mul(object_quat, math_utils.quat_conjugate(target_quat))
-    return 2.0 * torch.asin(torch.clamp(torch.linalg.norm(quat_diff[:, 0:3], ord=2, dim=-1), max=1.0))
-
-
 @torch.jit.script
 def evaluate_reorient_success(
     object_quat: torch.Tensor, target_quat: torch.Tensor, success_tolerance: float
@@ -173,12 +158,12 @@ def evaluate_reorient_success(
     Returns:
         Per-environment success flags and orientation errors [rad].
     """
-    orientation_error = direct_reorient_rotation_distance(object_quat, target_quat)
+    orientation_error = math_utils.quat_error_magnitude(object_quat, target_quat)
     return orientation_error <= success_tolerance, orientation_error
 
 
 @torch.jit.script
-def direct_reorient_reward(
+def reorient_reward(
     reset_buf: torch.Tensor,
     reset_goal_buf: torch.Tensor,
     successes: torch.Tensor,
@@ -310,7 +295,7 @@ def __call__(
             asset.data.root_quat_w.torch, command[:, 3:7], success_tolerance
         )
         self._orientation_error.update(orientation_error)
-        reward, _, self._successes, self._consecutive_successes = direct_reorient_reward(
+        reward, _, self._successes, self._consecutive_successes = reorient_reward(
             env.reset_buf,
             torch.zeros_like(env.reset_buf),
             self._successes,

source/isaaclab_tasks/isaaclab_tasks/core/reorient/reorient_direct_env.py

@@ -20,7 +20,7 @@
 from isaaclab.sim.spawners.from_files import GroundPlaneCfg, spawn_ground_plane
 from isaaclab.utils.math import quat_conjugate, quat_mul, sample_uniform, saturate, scale_transform, unscale_transform
 
-from isaaclab_tasks.core.reorient.mdp.rewards import direct_reorient_reward, evaluate_reorient_success
+from isaaclab_tasks.core.reorient.mdp.rewards import evaluate_reorient_success, reorient_reward
 from isaaclab_tasks.core.reorient.reorient_task_base import GOAL_MARKER_POSITION, IN_HAND_POS_OFFSET
 from isaaclab_tasks.core.utils import EpisodeErrorRecorder, randomize_rotation, sample_joint_positions_within_limits
 
@@ -185,7 +185,7 @@ def _get_rewards(self) -> torch.Tensor:
         # the success flags and orientation errors were computed this step by
         # :meth:`_get_dones`; the recorder and the reward reuse them
         self._orientation_error.update(self._orientation_error_buf)
-        total_reward, goal_resets, successes, consecutive_successes = direct_reorient_reward(
+        total_reward, goal_resets, successes, consecutive_successes = reorient_reward(
             self.reset_buf,
             self.reset_goal_buf,
             self.successes,

source/isaaclab_tasks/test/core/test_dexterous_task_math.py

@@ -19,7 +19,7 @@
 
 from isaaclab_tasks.core.handover.mdp.rewards import evaluate_handover_success, handover_reward
 from isaaclab_tasks.core.reorient.mdp.observations import compute_cube_keypoints, cube_keypoints_from_quat
-from isaaclab_tasks.core.reorient.mdp.rewards import direct_reorient_rotation_distance, evaluate_reorient_success
+from isaaclab_tasks.core.reorient.mdp.rewards import evaluate_reorient_success
 
 _DEVICES = ["cpu"] + (["cuda:0"] if torch.cuda.is_available() else [])
 
@@ -37,8 +37,8 @@ def _quats(device, *quats):
 
 @pytest.mark.parametrize("device", _DEVICES)
 def test_rotation_distance_recovers_known_angles(device):
-    distance = direct_reorient_rotation_distance(
-        _quats(device, _IDENTITY, _ROT90_X, _ROT180_Z), _quats(device, _IDENTITY, _IDENTITY, _IDENTITY)
+    _, distance = evaluate_reorient_success(
+        _quats(device, _IDENTITY, _ROT90_X, _ROT180_Z), _quats(device, _IDENTITY, _IDENTITY, _IDENTITY), 0.4
     )
     torch.testing.assert_close(distance, torch.tensor([0.0, math.pi / 2, math.pi], device=device), atol=1e-5, rtol=0.0)