Adds freeze option in NavigationSE2Action, improves observation MDPs (#7)

# Description

**Added**

- Added ``freeze_low_level_policy`` option to :`NavigationSE2ActionCfg`
to allow to not freeze the low level policy.
- Added ``nan_fill_value`` option to `camera_image` to allow to fill
NaNs with a specific value.

**Changed**

- Updated `height_scan_clipped` to make clipping optional.

## Type of change

- New feature (non-breaking change which adds functionality)

## Checklist

- [x] I have run the [`pre-commit` checks](https://pre-commit.com/) with
`./formatter.sh`
- [ ] I have made corresponding changes to the documentation (will come
shortly)
- [x] My changes generate no new warnings
- [x] I have updated the changelog and the corresponding version in the
extension's `config/extension.toml` file
- [x] I have added my name to the `CONTRIBUTORS.md` or my name already
exists there

Author: pascal-roth

exts/nav_tasks/config/extension.toml

@@ -1,7 +1,7 @@
 [package]
 
 # Note: Semantic Versioning is used: https://semver.org/
-version = "0.3.10"
+version = "0.3.11"
 
 # Description
 title = "IsaacLab Navigation RL Tasks"

exts/nav_tasks/docs/CHANGELOG.rst

@@ -1,6 +1,22 @@
 Changelog
 ---------
 
+
+0.3.11 (2025-08-13)
+~~~~~~~~~~~~~~~~~~
+
+Added
+^^^^^
+
+- Added ``freeze_low_level_policy`` option to :class:`nav_tasks.mdp.actions.NavigationSE2ActionCfg` to allow to not freeze the low level policy.
+- Added ``nan_fill_value`` option to :func:`nav_tasks.mdp.observations.camera_observations:camera_image` to allow to fill NaNs with a specific value.
+
+Changed
+^^^^^^^
+
+- Updated :func:`nav_tasks.mdp.observations.height_scan_observations:height_scan_clipped` to make clipping optional.
+
+
 0.3.10 (2025-08-08)
 ~~~~~~~~~~~~~~~~~~
 

exts/nav_tasks/nav_tasks/mdp/actions/navigation_actions.py

@@ -30,7 +30,9 @@ def __init__(self, cfg: NavigationSE2ActionCfg, env: ManagerBasedRLEnv):
         # load policies
         file_bytes = read_file(self.cfg.low_level_policy_file)
         self.low_level_policy = torch.jit.load(file_bytes, map_location=self.device)
-        self.low_level_policy = torch.jit.freeze(self.low_level_policy.eval())
+        self.low_level_policy.eval()
+        if self.cfg.freeze_low_level_policy:
+            self.low_level_policy = torch.jit.freeze(self.low_level_policy)
 
         # prepare joint position actions
         if not isinstance(self.cfg.low_level_action, list):

exts/nav_tasks/nav_tasks/mdp/actions/navigation_actions_cfg.py

@@ -28,6 +28,11 @@ class NavigationSE2ActionCfg(ActionTermCfg):
     low_level_policy_file: str = MISSING
     """Path to the low level policy file."""
 
+    freeze_low_level_policy: bool = True
+    """Whether to freeze the low level policy.
+
+    Can improve performance but will also eliminate possible functions such as `reset`."""
+
     low_level_obs_group: str = "low_level_policy"
     """Observation group of the low level policy."""
 

exts/nav_tasks/nav_tasks/mdp/observations/camera_observations.py

@@ -22,7 +22,11 @@
 
 
 def camera_image(
-    env: ManagerBasedEnv, sensor_cfg: SceneEntityCfg, data_type: str = "distance_to_image_plane", flatten: bool = False
+    env: ManagerBasedEnv,
+    sensor_cfg: SceneEntityCfg,
+    data_type: str = "distance_to_image_plane",
+    flatten: bool = False,
+    nan_fill_value: float | None = None,
 ) -> torch.Tensor:
     """Camera image Observations.
 
@@ -34,6 +38,7 @@ def camera_image(
         sensor_cfg: The name of the sensor.
         data_type: The type of data to extract from the sensor. Default is "distance_to_image_plane".
         flatten: If True, the image will be flattened to 1D. Default is False.
+        nan_fill_value: The value to fill nan/inf values with. If None, the maximum distance of the sensor will be used.
 
     Returns:
         The image data."""
@@ -43,8 +48,9 @@ def camera_image(
     img = sensor.data.output[data_type].clone()
 
     if data_type == "distance_to_image_plane":
-        img[torch.isnan(img)] = sensor.cfg.max_distance
-        img[torch.isinf(img)] = sensor.cfg.max_distance
+        if nan_fill_value is None:
+            nan_fill_value = sensor.cfg.max_distance
+        img = torch.nan_to_num(img, nan=nan_fill_value, posinf=nan_fill_value, neginf=0.0)
 
     # if type torch.uint8, convert to float and scale between 0 and 1
     if img.dtype == torch.uint8:

exts/nav_tasks/nav_tasks/mdp/observations/height_scan_observations.py

@@ -49,16 +49,15 @@ def height_scan_clipped(
     env: ManagerBasedRLEnv,
     sensor_cfg: SceneEntityCfg,
     offset: float = 0.5,
-    clip_height: tuple[float, float] = (-1.0, 0.5),
+    clip_height: tuple[float, float] | None = (-1.0, 0.5),
 ) -> torch.Tensor:
-    """Height scan from the given sensor w.r.t. the sensor's frame.
-
-    The provided offset (Defaults to 0.5) is subtracted from the returned values.
-    """
+    """Height scan from the given sensor w.r.t. the sensor's frame."""
     # get the bounded height scan
     height = height_scan_bounded(env, sensor_cfg, offset)
-    # clip to max observable height
-    return torch.clip(height, clip_height[0], clip_height[1])
+    if clip_height is not None:
+        # clip to max observable height
+        height = torch.clip(height, clip_height[0], clip_height[1])
+    return height
 
 
 def height_scan_square(