Adds option to define the concatenation dimension in the ObservationManager and change counter update in CommandManager (#2393)

# Description


Added support for concatenation of observations along different
dimensions in `ObservationManager`.

Updates the position where the command counter is increased to allow
checking for reset environments in the resample call of the
`CommandManager`

## 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
`./isaaclab.sh --format`
- [ ] I have made corresponding changes to the documentation
- [x] My changes generate no new warnings
- [x] I have added tests that prove my fix is effective or that my
feature works
- [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

---------

Signed-off-by: Pascal Roth <[email protected]>
Signed-off-by: Kelly Guo <[email protected]>
Signed-off-by: Kelly Guo <[email protected]>
Co-authored-by: Mayank Mittal <[email protected]>
Co-authored-by: Kelly Guo <[email protected]>
Co-authored-by: Kelly Guo <[email protected]>

Author: 3 people

source/isaaclab/config/extension.toml

@@ -1,7 +1,7 @@
 [package]
 
 # Note: Semantic Versioning is used: https://semver.org/
-version = "0.39.4"
+version = "0.39.5"
 
 # Description
 title = "Isaac Lab framework for Robot Learning"

source/isaaclab/docs/CHANGELOG.rst

@@ -1,6 +1,21 @@
 Changelog
 ---------
 
+0.39.5 (2025-05-16)
+~~~~~~~~~~~~~~~~~~~
+
+Added
+^^^^^
+
+* Added support for concatenation of observations along different dimensions in :class:`~isaaclab.managers.observation_manager.ObservationManager`.
+
+Changed
+^^^^^^^
+
+* Updated the :class:`~isaaclab.managers.command_manager.CommandManager` to update the command counter after the
+  resampling call.
+
+
 0.39.4 (2025-05-16)
 ~~~~~~~~~~~~~~~~~~~
 

source/isaaclab/isaaclab/managers/command_manager.py

@@ -181,10 +181,10 @@ def _resample(self, env_ids: Sequence[int]):
         if len(env_ids) != 0:
             # resample the time left before resampling
             self.time_left[env_ids] = self.time_left[env_ids].uniform_(*self.cfg.resampling_time_range)
-            # increment the command counter
-            self.command_counter[env_ids] += 1
             # resample the command
             self._resample_command(env_ids)
+            # increment the command counter
+            self.command_counter[env_ids] += 1
 
     """
     Implementation specific functions.

source/isaaclab/isaaclab/managers/manager_term_cfg.py

@@ -201,12 +201,22 @@ class ObservationGroupCfg:
     concatenate_terms: bool = True
     """Whether to concatenate the observation terms in the group. Defaults to True.
 
-    If true, the observation terms in the group are concatenated along the last dimension.
+    If true, the observation terms in the group are concatenated along the dimension specified through :attr:`concatenate_dim`.
     Otherwise, they are kept separate and returned as a dictionary.
 
     If the observation group contains terms of different dimensions, it must be set to False.
     """
 
+    concatenate_dim: int = -1
+    """Dimension along to concatenate the different observation terms. Defaults to -1, which
+    means the last dimension of the observation terms.
+
+    If :attr:`concatenate_terms` is True, this parameter specifies the dimension along which the observation terms are concatenated.
+    The indicated dimension depends on the shape of the observations. For instance, for a 2D RGB image of shape (H, W, C), the dimension
+    0 means concatenating along the height, 1 along the width, and 2 along the channels. The offset due
+    to the batched environment is handled automatically.
+    """
+
     enable_corruption: bool = False
     """Whether to enable corruption for the observation group. Defaults to False.
 

source/isaaclab/isaaclab/managers/observation_manager.py

@@ -88,8 +88,18 @@ def __init__(self, cfg: object, env: ManagerBasedEnv):
             # otherwise, keep the list of shapes as is
             if self._group_obs_concatenate[group_name]:
                 try:
-                    term_dims = [torch.tensor(dims, device="cpu") for dims in group_term_dims]
-                    self._group_obs_dim[group_name] = tuple(torch.sum(torch.stack(term_dims, dim=0), dim=0).tolist())
+                    term_dims = torch.stack([torch.tensor(dims, device="cpu") for dims in group_term_dims], dim=0)
+                    if len(term_dims.shape) > 1:
+                        if self._group_obs_concatenate_dim[group_name] >= 0:
+                            dim = self._group_obs_concatenate_dim[group_name] - 1  # account for the batch offset
+                        else:
+                            dim = self._group_obs_concatenate_dim[group_name]
+                        dim_sum = torch.sum(term_dims[:, dim], dim=0)
+                        term_dims[0, dim] = dim_sum
+                        term_dims = term_dims[0]
+                    else:
+                        term_dims = torch.sum(term_dims, dim=0)
+                    self._group_obs_dim[group_name] = tuple(term_dims.tolist())
                 except RuntimeError:
                     raise RuntimeError(
                         f"Unable to concatenate observation terms in group '{group_name}'."
@@ -330,7 +340,8 @@ def compute_group(self, group_name: str) -> torch.Tensor | dict[str, torch.Tenso
 
         # concatenate all observations in the group together
         if self._group_obs_concatenate[group_name]:
-            return torch.cat(list(group_obs.values()), dim=-1)
+            # set the concatenate dimension, account for the batch dimension if positive dimension is given
+            return torch.cat(list(group_obs.values()), dim=self._group_obs_concatenate_dim[group_name])
         else:
             return group_obs
 
@@ -370,6 +381,8 @@ def _prepare_terms(self):
         self._group_obs_term_cfgs: dict[str, list[ObservationTermCfg]] = dict()
         self._group_obs_class_term_cfgs: dict[str, list[ObservationTermCfg]] = dict()
         self._group_obs_concatenate: dict[str, bool] = dict()
+        self._group_obs_concatenate_dim: dict[str, int] = dict()
+
         self._group_obs_term_history_buffer: dict[str, dict] = dict()
         # create a list to store modifiers that are classes
         # we store it as a separate list to only call reset on them and prevent unnecessary calls
@@ -407,6 +420,9 @@ def _prepare_terms(self):
             group_entry_history_buffer: dict[str, CircularBuffer] = dict()
             # read common config for the group
             self._group_obs_concatenate[group_name] = group_cfg.concatenate_terms
+            self._group_obs_concatenate_dim[group_name] = (
+                group_cfg.concatenate_dim + 1 if group_cfg.concatenate_dim >= 0 else group_cfg.concatenate_dim
+            )
             # check if config is dict already
             if isinstance(group_cfg, dict):
                 group_cfg_items = group_cfg.items()
@@ -415,7 +431,13 @@ def _prepare_terms(self):
             # iterate over all the terms in each group
             for term_name, term_cfg in group_cfg_items:
                 # skip non-obs settings
-                if term_name in ["enable_corruption", "concatenate_terms", "history_length", "flatten_history_dim"]:
+                if term_name in [
+                    "enable_corruption",
+                    "concatenate_terms",
+                    "history_length",
+                    "flatten_history_dim",
+                    "concatenate_dim",
+                ]:
                     continue
                 # check for non config
                 if term_cfg is None:

source/isaaclab/test/managers/test_observation_manager.py

@@ -667,41 +667,43 @@ class CriticCfg(ObservationGroupCfg):
     assert torch.min(obs_critic["term_4"]) >= -0.5
     assert torch.max(obs_critic["term_4"]) <= 0.5
 
-    def test_serialize(self):
-        """Test serialize call for ManagerTermBase terms."""
 
-        serialize_data = {"test": 0}
+def test_serialize(setup_env):
+    """Test serialize call for ManagerTermBase terms."""
+    env = setup_env
 
-        class test_serialize_term(ManagerTermBase):
+    serialize_data = {"test": 0}
 
-            def __init__(self, cfg: RewardTermCfg, env: ManagerBasedEnv):
-                super().__init__(cfg, env)
+    class test_serialize_term(ManagerTermBase):
 
-            def __call__(self, env: ManagerBasedEnv) -> torch.Tensor:
-                return grilled_chicken(env)
+        def __init__(self, cfg: RewardTermCfg, env: ManagerBasedEnv):
+            super().__init__(cfg, env)
 
-            def serialize(self) -> dict:
-                return serialize_data
+        def __call__(self, env: ManagerBasedEnv) -> torch.Tensor:
+            return grilled_chicken(env)
 
-        @configclass
-        class MyObservationManagerCfg:
-            """Test config class for observation manager."""
+        def serialize(self) -> dict:
+            return serialize_data
+
+    @configclass
+    class MyObservationManagerCfg:
+        """Test config class for observation manager."""
 
-            @configclass
-            class PolicyCfg(ObservationGroupCfg):
-                """Test config class for policy observation group."""
+        @configclass
+        class PolicyCfg(ObservationGroupCfg):
+            """Test config class for policy observation group."""
 
-                concatenate_terms = False
-                term_1 = ObservationTermCfg(func=test_serialize_term)
+            concatenate_terms = False
+            term_1 = ObservationTermCfg(func=test_serialize_term)
 
-            policy: ObservationGroupCfg = PolicyCfg()
+        policy: ObservationGroupCfg = PolicyCfg()
 
-        # create observation manager
-        cfg = MyObservationManagerCfg()
-        self.obs_man = ObservationManager(cfg, self.env)
+    # create observation manager
+    cfg = MyObservationManagerCfg()
+    obs_man = ObservationManager(cfg, env)
 
-        # check expected output
-        self.assertEqual(self.obs_man.serialize(), {"policy": {"term_1": serialize_data}})
+    # check expected output
+    assert obs_man.serialize() == {"policy": {"term_1": serialize_data}}
 
 
 def test_modifier_invalid_config(setup_env):
@@ -728,3 +730,71 @@ class PolicyCfg(ObservationGroupCfg):
 
     with pytest.raises(ValueError):
         ObservationManager(cfg, env)
+
+
+def test_concatenate_dim(setup_env):
+    """Test concatenation of observations along different dimensions."""
+    env = setup_env
+
+    @configclass
+    class MyObservationManagerCfg:
+        """Test config class for observation manager."""
+
+        @configclass
+        class PolicyCfg(ObservationGroupCfg):
+            """Test config class for policy observation group."""
+
+            concatenate_terms = True
+            concatenate_dim = 1  # Concatenate along dimension 1
+            term_1 = ObservationTermCfg(func=grilled_chicken_image, scale=1.0, params={"bland": 1.0, "channel": 1})
+            term_2 = ObservationTermCfg(func=grilled_chicken_image, scale=1.0, params={"bland": 1.0, "channel": 1})
+
+        @configclass
+        class CriticCfg(ObservationGroupCfg):
+            """Test config class for critic observation group."""
+
+            concatenate_terms = True
+            concatenate_dim = 2  # Concatenate along dimension 2
+            term_1 = ObservationTermCfg(func=grilled_chicken_image, scale=1.0, params={"bland": 1.0, "channel": 1})
+            term_2 = ObservationTermCfg(func=grilled_chicken_image, scale=1.0, params={"bland": 1.0, "channel": 1})
+
+        @configclass
+        class CriticCfg_neg_dim(ObservationGroupCfg):
+            """Test config class for critic observation group."""
+
+            concatenate_terms = True
+            concatenate_dim = -1  # Concatenate along last dimension
+            term_1 = ObservationTermCfg(func=grilled_chicken_image, scale=1.0, params={"bland": 1.0, "channel": 1})
+            term_2 = ObservationTermCfg(func=grilled_chicken_image, scale=1.0, params={"bland": 1.0, "channel": 1})
+
+        policy: ObservationGroupCfg = PolicyCfg()
+        critic: ObservationGroupCfg = CriticCfg()
+        critic_neg_dim: ObservationGroupCfg = CriticCfg_neg_dim()
+
+    # create observation manager
+    cfg = MyObservationManagerCfg()
+    obs_man = ObservationManager(cfg, env)
+    # compute observation using manager
+    observations = obs_man.compute()
+
+    # obtain the group observations
+    obs_policy: torch.Tensor = observations["policy"]
+    obs_critic: torch.Tensor = observations["critic"]
+    obs_critic_neg_dim: torch.Tensor = observations["critic_neg_dim"]
+
+    # check the observation shapes
+    # For policy: concatenated along dim 1, so width should be doubled
+    assert obs_policy.shape == (env.num_envs, 128, 512, 1)
+    # For critic: concatenated along last dim, so channels should be doubled
+    assert obs_critic.shape == (env.num_envs, 128, 256, 2)
+    # For critic_neg_dim: concatenated along last dim, so channels should be doubled
+    assert obs_critic_neg_dim.shape == (env.num_envs, 128, 256, 2)
+
+    # verify the data is concatenated correctly
+    # For policy: check that the second half matches the first half
+    torch.testing.assert_close(obs_policy[:, :, :256, :], obs_policy[:, :, 256:, :])
+    # For critic: check that the second channel matches the first channel
+    torch.testing.assert_close(obs_critic[:, :, :, 0], obs_critic[:, :, :, 1])
+
+    # For critic_neg_dim: check that it is the same as critic
+    torch.testing.assert_close(obs_critic_neg_dim, obs_critic)