diff --git a/rllib/connectors/common/add_observations_from_episodes_to_batch.py b/rllib/connectors/common/add_observations_from_episodes_to_batch.py
index 9f5aad6f1c47..7d55251a841e 100644
--- a/rllib/connectors/common/add_observations_from_episodes_to_batch.py
+++ b/rllib/connectors/common/add_observations_from_episodes_to_batch.py
@@ -7,11 +7,40 @@
 from ray.rllib.core.rl_module.rl_module import RLModule
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class AddObservationsFromEpisodesToBatch(ConnectorV2):
     """Gets the last observation from a running episode and adds it to the batch.
 
+    Note: This is one of the default env-to-module or Learner ConnectorV2 pieces that
+    are added automatically by RLlib into every env-to-module/Learner connector
+    pipeline, unless `config.add_default_connectors_to_env_to_module_pipeline` or
+    `config.add_default_connectors_to_learner_pipeline ` are set to
+    False.
+
+    The default env-to-module connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+    The default Learner connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddColumnsFromEpisodesToTrainBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+
+    This ConnectorV2:
     - Operates on a list of Episode objects.
     - Gets the most recent observation(s) from all the given episodes and adds them
     to the batch under construction (as a list of individual observations).
diff --git a/rllib/connectors/common/add_states_from_episodes_to_batch.py b/rllib/connectors/common/add_states_from_episodes_to_batch.py
index 9f6e602387ae..4b15229093b6 100644
--- a/rllib/connectors/common/add_states_from_episodes_to_batch.py
+++ b/rllib/connectors/common/add_states_from_episodes_to_batch.py
@@ -15,11 +15,39 @@
 from ray.rllib.utils.numpy import convert_to_numpy
 from ray.rllib.utils.spaces.space_utils import batch, BatchedNdArray
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class AddStatesFromEpisodesToBatch(ConnectorV2):
     """Gets last STATE_OUT from running episode and adds it as STATE_IN to the batch.
 
+    Note: This is one of the default env-to-module or Learner ConnectorV2 pieces that
+    are added automatically by RLlib into every env-to-module/Learner connector
+    pipeline, unless `config.add_default_connectors_to_env_to_module_pipeline` or
+    `config.add_default_connectors_to_learner_pipeline ` are set to
+    False.
+
+    The default env-to-module connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+    The default Learner connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddColumnsFromEpisodesToTrainBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+
     If the RLModule is stateful, the episodes' STATE_OUTS will be extracted
     and restructured under a new STATE_IN key.
     As a Learner connector, the resulting STATE_IN batch has the shape (B', ...).
@@ -32,6 +60,7 @@ class AddStatesFromEpisodesToBatch(ConnectorV2):
     reshaped into (B, T=max_seq_len (learner) or 1 (env-to-module), ...) and will be
     zero-padded, if necessary.
 
+    This ConnectorV2:
     - Operates on a list of Episode objects.
     - Gets the most recent STATE_OUT from all the given episodes and adds them under
     the STATE_IN key to the batch under construction.
diff --git a/rllib/connectors/common/agent_to_module_mapping.py b/rllib/connectors/common/agent_to_module_mapping.py
index ee6a738ee41e..1f8a140caf9f 100644
--- a/rllib/connectors/common/agent_to_module_mapping.py
+++ b/rllib/connectors/common/agent_to_module_mapping.py
@@ -8,11 +8,39 @@
 from ray.rllib.env.multi_agent_episode import MultiAgentEpisode
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.typing import EpisodeType, ModuleID
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class AgentToModuleMapping(ConnectorV2):
     """ConnectorV2 that performs mapping of data from AgentID based to ModuleID based.
 
+    Note: This is one of the default env-to-module or Learner ConnectorV2 pieces that
+    are added automatically by RLlib into every env-to-module/Learner connector
+    pipeline, unless `config.add_default_connectors_to_env_to_module_pipeline` or
+    `config.add_default_connectors_to_learner_pipeline ` are set to
+    False.
+
+    The default env-to-module connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+    The default Learner connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddColumnsFromEpisodesToTrainBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+
     This connector piece is only used by RLlib (as a default connector piece) in a
     multi-agent setup.
 
diff --git a/rllib/connectors/common/batch_individual_items.py b/rllib/connectors/common/batch_individual_items.py
index 02b72c870b2a..f7c5c49eb5ce 100644
--- a/rllib/connectors/common/batch_individual_items.py
+++ b/rllib/connectors/common/batch_individual_items.py
@@ -10,9 +10,93 @@
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.spaces.space_utils import batch
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class BatchIndividualItems(ConnectorV2):
+    """Batches individual data-items (in lists) into tensors (with batch dimension).
+
+    Note: This is one of the default env-to-module or Learner ConnectorV2 pieces that
+    are added automatically by RLlib into every env-to-module/Learner connector
+    pipeline, unless `config.add_default_connectors_to_env_to_module_pipeline` or
+    `config.add_default_connectors_to_learner_pipeline ` are set to
+    False.
+
+    The default env-to-module connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+    The default Learner connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddColumnsFromEpisodesToTrainBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+
+    This ConnectorV2:
+    - Operates only on the input `data`, NOT the incoming list of episode objects
+    (ignored).
+    - In the single-agent case, `data` must already be a dict, structured as follows by
+    prior connector pieces of the same pipeline:
+    [col0] -> {[(eps_id,)]: [list of individual batch items]}
+    - In the multi-agent case, `data` must already be a dict, structured as follows by
+    prior connector pieces of the same pipeline (in particular the
+    `AgentToModuleMapping` piece):
+    [module_id] -> [col0] -> [list of individual batch items]
+    - Translates the above data under the different columns (e.g. "obs") into final
+    (batched) structures. For the single-agent case, the output `data` looks like this:
+    [col0] -> [possibly complex struct of batches (at the leafs)].
+    For the multi-agent case, the output `data` looks like this:
+    [module_id] -> [col0] -> [possibly complex struct of batches (at the leafs)].
+
+    .. testcode::
+
+        from ray.rllib.connectors.common import BatchIndividualItems
+        from ray.rllib.utils.test_utils import check
+
+        single_agent_data = {
+            "obs": {
+                # Note that at this stage, next-obs is not part of the data anymore ..
+                ("MA-EPS0",): [0, 1],
+                ("MA-EPS1",): [2, 3],
+            },
+            "actions": {
+                # .. so we have as many actions per episode as we have observations.
+                ("MA-EPS0",): [4, 5],
+                ("MA-EPS1",): [6, 7],
+            },
+        }
+
+        # Create our (single-agent) connector piece.
+        connector = BatchIndividualItems()
+
+        # Call the connector (and thereby batch the individual items).
+        output_data = connector(
+            rl_module=None,  # This particular connector works without an RLModule.
+            data=single_agent_data,
+            episodes=[],  # This particular connector works without a list of episodes.
+            explore=True,
+            shared_data={},
+        )
+
+        # `data` should now be batched (episode IDs should have been removed from the
+        # struct).
+        check(
+            output_data,
+            {"obs": [0, 1, 2, 3], "actions": [4, 5, 6, 7]},
+        )
+    """
+
     def __init__(
         self,
         input_observation_space: Optional[gym.Space] = None,
@@ -52,7 +136,7 @@ def __call__(
             # Multi-agent case: This connector piece should only be used after(!)
             # the AgentToModuleMapping connector has already been applied, leading
             # to a batch structure of:
-            # [module_id] -> [col0] -> [list of items]
+            # [module_id] -> [col0] -> [list of individual batch items]
             if is_multi_rl_module and column in rl_module:
                 # Case, in which a column has already been properly batched before this
                 # connector piece is called.
diff --git a/rllib/connectors/common/module_to_agent_unmapping.py b/rllib/connectors/common/module_to_agent_unmapping.py
index c17854db1ede..399d80d09914 100644
--- a/rllib/connectors/common/module_to_agent_unmapping.py
+++ b/rllib/connectors/common/module_to_agent_unmapping.py
@@ -6,8 +6,10 @@
 from ray.rllib.env.multi_agent_episode import MultiAgentEpisode
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class ModuleToAgentUnmapping(ConnectorV2):
     """Performs flipping of `data` from ModuleID- to AgentID based mapping.
 
diff --git a/rllib/connectors/common/numpy_to_tensor.py b/rllib/connectors/common/numpy_to_tensor.py
index 78cb9a02ad39..d6aa689e38a1 100644
--- a/rllib/connectors/common/numpy_to_tensor.py
+++ b/rllib/connectors/common/numpy_to_tensor.py
@@ -10,13 +10,45 @@
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.torch_utils import convert_to_torch_tensor
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class NumpyToTensor(ConnectorV2):
     """Converts numpy arrays across the entire input data into (framework) tensors.
 
     The framework information is received via the provided `rl_module` arg in the
-    `__call__`.
+    `__call__()` method.
+
+    Note: This is one of the default env-to-module or Learner ConnectorV2 pieces that
+    are added automatically by RLlib into every env-to-module/Learner connector
+    pipeline, unless `config.add_default_connectors_to_env_to_module_pipeline` or
+    `config.add_default_connectors_to_learner_pipeline ` are set to
+    False.
+
+    The default env-to-module connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+    The default Learner connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddColumnsFromEpisodesToTrainBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+
+    This ConnectorV2:
+    - Loops through the input `data` and converts all found numpy arrays into
+    framework-specific tensors (possibly on a GPU).
     """
 
     def __init__(
@@ -37,6 +69,8 @@ def __init__(
             pin_mempory: Whether to pin memory when creating (torch) tensors.
                 If None (default), pins memory if `as_learner_connector` is True,
                 otherwise doesn't pin memory.
+            device: An optional device to move the resulting tensors to. If not
+                provided, all data will be left on the CPU.
             **kwargs:
         """
         super().__init__(
@@ -68,7 +102,6 @@ def __call__(
             is_single_agent = True
             data = {DEFAULT_MODULE_ID: data}
 
-        # TODO (sven): Support specifying a device (e.g. GPU).
         for module_id, module_data in data.copy().items():
             infos = module_data.pop(Columns.INFOS, None)
             if rl_module.framework == "torch":
diff --git a/rllib/connectors/common/tensor_to_numpy.py b/rllib/connectors/common/tensor_to_numpy.py
index d3c6f9b390b2..f2def9f8566c 100644
--- a/rllib/connectors/common/tensor_to_numpy.py
+++ b/rllib/connectors/common/tensor_to_numpy.py
@@ -5,8 +5,10 @@
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.numpy import convert_to_numpy
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class TensorToNumpy(ConnectorV2):
     """Converts (framework) tensors across the entire input data into numpy arrays."""
 
diff --git a/rllib/connectors/env_to_module/mean_std_filter.py b/rllib/connectors/env_to_module/mean_std_filter.py
index ad583ed530d3..d568a7bc36a4 100644
--- a/rllib/connectors/env_to_module/mean_std_filter.py
+++ b/rllib/connectors/env_to_module/mean_std_filter.py
@@ -8,11 +8,10 @@
 from ray.rllib.connectors.connector_v2 import ConnectorV2
 from ray.rllib.core.rl_module.rl_module import RLModule
 from ray.rllib.utils.annotations import override
-from ray.rllib.utils.filter import MeanStdFilter as _MeanStdFilter
+from ray.rllib.utils.filter import MeanStdFilter as _MeanStdFilter, RunningStat
 from ray.rllib.utils.spaces.space_utils import get_base_struct_from_space
 from ray.rllib.utils.typing import AgentID, EpisodeType, StateDict
 from ray.util.annotations import PublicAPI
-from ray.rllib.utils.filter import RunningStat
 
 
 @PublicAPI(stability="alpha")
diff --git a/rllib/connectors/env_to_module/observation_preprocessor.py b/rllib/connectors/env_to_module/observation_preprocessor.py
new file mode 100644
index 000000000000..9278df42bd45
--- /dev/null
+++ b/rllib/connectors/env_to_module/observation_preprocessor.py
@@ -0,0 +1,72 @@
+import abc
+from typing import Any, List, Optional
+
+from ray.rllib.connectors.connector_v2 import ConnectorV2
+from ray.rllib.core.rl_module.rl_module import RLModule
+from ray.rllib.utils.annotations import override
+from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
+
+
+@PublicAPI(stability="alpha")
+class ObservationPreprocessor(ConnectorV2, abc.ABC):
+    """Env-to-module connector performing one preprocessor step on the last observation.
+
+    This is a convenience class that simplifies the writing of few-step preprocessor
+    connectors.
+
+    Users must implement the `preprocess()` method, which simplifies the usual procedure
+    of extracting some data from a list of episodes and adding it to the batch to a mere
+    "old-observation --transform--> return new-observation" step.
+    """
+
+    @override(ConnectorV2)
+    def recompute_observation_space_from_input_spaces(self):
+        # Users should override this method only in case the `ObservationPreprocessor`
+        # changes the observation space of the pipeline. In this case, return the new
+        # observation space based on the incoming one (`self.input_observation_space`).
+        super().recompute_observation_space_from_input_spaces()
+
+    @abc.abstractmethod
+    def preprocess(self, observation):
+        """Override to implement the preprocessing logic.
+
+        Args:
+            observation: A single (non-batched) observation item for a single agent to
+                be processed by this connector.
+
+        Returns:
+            The new observation after `observation` has been preprocessed.
+        """
+
+    @override(ConnectorV2)
+    def __call__(
+        self,
+        *,
+        rl_module: RLModule,
+        data: Any,
+        episodes: List[EpisodeType],
+        explore: Optional[bool] = None,
+        persistent_data: Optional[dict] = None,
+        **kwargs,
+    ) -> Any:
+        # We process and then replace observations inside the episodes directly.
+        # Thus, all following connectors will only see and operate on the already
+        # processed observation (w/o having access anymore to the original
+        # observations).
+        for sa_episode in self.single_agent_episode_iterator(episodes):
+            observation = sa_episode.get_observations(-1)
+
+            # Process the observation and write the new observation back into the
+            # episode.
+            new_observation = self.preprocess(observation=observation)
+            sa_episode.set_observations(at_indices=-1, new_data=new_observation)
+            #  We set the Episode's observation space to ours so that we can safely
+            #  set the last obs to the new value (without causing a space mismatch
+            #  error).
+            sa_episode.observation_space = self.observation_space
+
+        # Leave `data` as is. RLlib's default connector will automatically
+        # populate the OBS column therein from the episodes' now transformed
+        # observations.
+        return data
diff --git a/rllib/connectors/env_to_module/prev_actions_prev_rewards.py b/rllib/connectors/env_to_module/prev_actions_prev_rewards.py
index 5b26cd1f8b87..2ba089c5f91c 100644
--- a/rllib/connectors/env_to_module/prev_actions_prev_rewards.py
+++ b/rllib/connectors/env_to_module/prev_actions_prev_rewards.py
@@ -9,8 +9,10 @@
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.spaces.space_utils import batch, flatten_to_single_ndarray
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class PrevActionsPrevRewards(ConnectorV2):
     """A connector piece that adds previous rewards and actions to the input obs.
 
diff --git a/rllib/connectors/env_to_module/write_observations_to_episodes.py b/rllib/connectors/env_to_module/write_observations_to_episodes.py
index f59fcf356151..7e460014e318 100644
--- a/rllib/connectors/env_to_module/write_observations_to_episodes.py
+++ b/rllib/connectors/env_to_module/write_observations_to_episodes.py
@@ -12,6 +12,21 @@
 class WriteObservationsToEpisodes(ConnectorV2):
     """Writes the observations from the batch into the running episodes.
 
+    Note: This is one of the default env-to-module ConnectorV2 pieces that are added
+    automatically by RLlib into every env-to-module connector pipelines, unless
+    `config.add_default_connectors_to_env_to_module_pipeline` is set to False.
+
+    The default env-to-module connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+
+    This ConnectorV2:
     - Operates on a batch that already has observations in it and a list of Episode
     objects.
     - Writes the observation(s) from the batch to all the given episodes. Thereby
diff --git a/rllib/connectors/learner/add_columns_from_episodes_to_train_batch.py b/rllib/connectors/learner/add_columns_from_episodes_to_train_batch.py
index 5c454ddcb647..1460bc8ae36f 100644
--- a/rllib/connectors/learner/add_columns_from_episodes_to_train_batch.py
+++ b/rllib/connectors/learner/add_columns_from_episodes_to_train_batch.py
@@ -5,13 +5,30 @@
 from ray.rllib.core.rl_module.rl_module import RLModule
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class AddColumnsFromEpisodesToTrainBatch(ConnectorV2):
     """Adds infos/actions/rewards/terminateds/... to train batch.
 
+    Note: This is one of the default Learner ConnectorV2 pieces that are added
+    automatically by RLlib into every Learner connector pipeline, unless
+    `config.add_default_connectors_to_learner_pipeline` is set to False.
+
+    The default Learner connector pipeline is:
+    [
+        [0 or more user defined ConnectorV2 pieces],
+        AddObservationsFromEpisodesToBatch,
+        AddColumnsFromEpisodesToTrainBatch,
+        AddStatesFromEpisodesToBatch,
+        AgentToModuleMapping,  # only in multi-agent setups!
+        BatchIndividualItems,
+        NumpyToTensor,
+    ]
+
     Does NOT add observations to train batch (these should have already been added
-    by a different ConnectorV2 piece: AddObservationsToTrainBatch)
+    by another ConnectorV2 piece: `AddObservationsToTrainBatch` in the same pipeline).
 
     If provided with `episodes` data, this connector piece makes sure that the final
     train batch going into the RLModule for updating (`forward_train()` call) contains
diff --git a/rllib/connectors/learner/add_next_observations_from_episodes_to_train_batch.py b/rllib/connectors/learner/add_next_observations_from_episodes_to_train_batch.py
index 4812ca43c524..fff6d57d714c 100644
--- a/rllib/connectors/learner/add_next_observations_from_episodes_to_train_batch.py
+++ b/rllib/connectors/learner/add_next_observations_from_episodes_to_train_batch.py
@@ -7,8 +7,10 @@
 from ray.rllib.core.rl_module.rl_module import RLModule
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class AddNextObservationsFromEpisodesToTrainBatch(ConnectorV2):
     """Adds the NEXT_OBS column with the correct episode observations to train batch.
 
diff --git a/rllib/connectors/learner/add_one_ts_to_episodes_and_truncate.py b/rllib/connectors/learner/add_one_ts_to_episodes_and_truncate.py
index 9e1cd68a88a3..0a408ed9de8c 100644
--- a/rllib/connectors/learner/add_one_ts_to_episodes_and_truncate.py
+++ b/rllib/connectors/learner/add_one_ts_to_episodes_and_truncate.py
@@ -6,8 +6,10 @@
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.postprocessing.episodes import add_one_ts_to_episodes_and_truncate
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class AddOneTsToEpisodesAndTruncate(ConnectorV2):
     """Adds an artificial timestep to all incoming episodes at the end.
 
diff --git a/rllib/connectors/module_to_env/get_actions.py b/rllib/connectors/module_to_env/get_actions.py
index cbade4e6c423..a54a230357c4 100644
--- a/rllib/connectors/module_to_env/get_actions.py
+++ b/rllib/connectors/module_to_env/get_actions.py
@@ -6,11 +6,32 @@
 from ray.rllib.env.multi_agent_episode import MultiAgentEpisode
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class GetActions(ConnectorV2):
     """Connector piece sampling actions from ACTION_DIST_INPUTS from an RLModule.
 
+    Note: This is one of the default module-to-env ConnectorV2 pieces that
+    are added automatically by RLlib into every module-to-env connector pipeline,
+    unless `config.add_default_connectors_to_module_to_env_pipeline` is set to
+    False.
+
+    The default module-to-env connector pipeline is:
+    [
+        GetActions,
+        TensorToNumpy,
+        UnBatchToIndividualItems,
+        ModuleToAgentUnmapping,  # only in multi-agent setups!
+        RemoveSingleTsTimeRankFromBatch,
+
+        [0 or more user defined ConnectorV2 pieces],
+
+        NormalizeAndClipActions,
+        ListifyDataForVectorEnv,
+    ]
+
     If necessary, this connector samples actions, given action dist. inputs and a
     dist. class.
     The connector will only sample from the action distribution, if the
diff --git a/rllib/connectors/module_to_env/listify_data_for_vector_env.py b/rllib/connectors/module_to_env/listify_data_for_vector_env.py
index cf042a0f556f..4c9d9e9ce807 100644
--- a/rllib/connectors/module_to_env/listify_data_for_vector_env.py
+++ b/rllib/connectors/module_to_env/listify_data_for_vector_env.py
@@ -7,11 +7,32 @@
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.spaces.space_utils import batch
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class ListifyDataForVectorEnv(ConnectorV2):
     """Performs conversion from ConnectorV2-style format to env/episode insertion.
 
+    Note: This is one of the default module-to-env ConnectorV2 pieces that
+    are added automatically by RLlib into every module-to-env connector pipeline,
+    unless `config.add_default_connectors_to_module_to_env_pipeline` is set to
+    False.
+
+    The default module-to-env connector pipeline is:
+    [
+        GetActions,
+        TensorToNumpy,
+        UnBatchToIndividualItems,
+        ModuleToAgentUnmapping,  # only in multi-agent setups!
+        RemoveSingleTsTimeRankFromBatch,
+
+        [0 or more user defined ConnectorV2 pieces],
+
+        NormalizeAndClipActions,
+        ListifyDataForVectorEnv,
+    ]
+
     Single agent case:
     Convert from:
     [col] -> [(env_vector_idx,)] -> [list of items].
diff --git a/rllib/connectors/module_to_env/normalize_and_clip_actions.py b/rllib/connectors/module_to_env/normalize_and_clip_actions.py
index 8a95fa725c4f..16168ffc0c40 100644
--- a/rllib/connectors/module_to_env/normalize_and_clip_actions.py
+++ b/rllib/connectors/module_to_env/normalize_and_clip_actions.py
@@ -13,9 +13,50 @@
     unsquash_action,
 )
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class NormalizeAndClipActions(ConnectorV2):
+    """Normalizes or clips actions in the input data (coming from the RLModule).
+
+    Note: This is one of the default module-to-env ConnectorV2 pieces that
+    are added automatically by RLlib into every module-to-env connector pipeline,
+    unless `config.add_default_connectors_to_module_to_env_pipeline` is set to
+    False.
+
+    The default module-to-env connector pipeline is:
+    [
+        GetActions,
+        TensorToNumpy,
+        UnBatchToIndividualItems,
+        ModuleToAgentUnmapping,  # only in multi-agent setups!
+        RemoveSingleTsTimeRankFromBatch,
+
+        [0 or more user defined ConnectorV2 pieces],
+
+        NormalizeAndClipActions,
+        ListifyDataForVectorEnv,
+    ]
+
+    This ConnectorV2:
+    - Deep copies the Columns.ACTIONS in the incoming `data` into a new column:
+    Columns.ACTIONS_FOR_ENV.
+    - Loops through the Columns.ACTIONS in the incoming `data` and normalizes or clips
+    these depending on the c'tor settings in `config.normalize_actions` and
+    `config.clip_actions`.
+    - Only applies to envs with Box action spaces.
+
+    Normalizing is the process of mapping NN-outputs (which are usually small
+    numbers, e.g. between -1.0 and 1.0) to the bounds defined by the action-space.
+    Normalizing helps the NN to learn faster in environments with large ranges between
+    `low` and `high` bounds or skewed action bounds (e.g. Box(-3000.0, 1.0, ...)).
+
+    Clipping clips the actions computed by the NN (and sampled from a distribution)
+    between the bounds defined by the action-space. Note that clipping is only performed
+    if `normalize_actions` is False.
+    """
+
     @override(ConnectorV2)
     def recompute_action_space_from_input_spaces(self) -> gym.Space:
         self._action_space_struct = get_base_struct_from_space(self.input_action_space)
@@ -52,9 +93,10 @@ def __init__(
                 from the resulting distribution, then this 0.9 will be clipped to 0.5
                 to fit into the [-0.5 0.5] interval.
         """
+        self._action_space_struct = None
+
         super().__init__(input_observation_space, input_action_space, **kwargs)
 
-        self._action_space_struct = None
         self.normalize_actions = normalize_actions
         self.clip_actions = clip_actions
 
diff --git a/rllib/connectors/module_to_env/remove_single_ts_time_rank_from_batch.py b/rllib/connectors/module_to_env/remove_single_ts_time_rank_from_batch.py
index 16de99578afd..15aebf266053 100644
--- a/rllib/connectors/module_to_env/remove_single_ts_time_rank_from_batch.py
+++ b/rllib/connectors/module_to_env/remove_single_ts_time_rank_from_batch.py
@@ -8,9 +8,33 @@
 from ray.rllib.core.rl_module.rl_module import RLModule
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class RemoveSingleTsTimeRankFromBatch(ConnectorV2):
+    """
+    Note: This is one of the default module-to-env ConnectorV2 pieces that
+    are added automatically by RLlib into every module-to-env connector pipeline,
+    unless `config.add_default_connectors_to_module_to_env_pipeline` is set to
+    False.
+
+    The default module-to-env connector pipeline is:
+    [
+        GetActions,
+        TensorToNumpy,
+        UnBatchToIndividualItems,
+        ModuleToAgentUnmapping,  # only in multi-agent setups!
+        RemoveSingleTsTimeRankFromBatch,
+
+        [0 or more user defined ConnectorV2 pieces],
+
+        NormalizeAndClipActions,
+        ListifyDataForVectorEnv,
+    ]
+
+    """
+
     @override(ConnectorV2)
     def __call__(
         self,
diff --git a/rllib/connectors/module_to_env/unbatch_to_individual_items.py b/rllib/connectors/module_to_env/unbatch_to_individual_items.py
index c992380adc29..5c271c5d9c78 100644
--- a/rllib/connectors/module_to_env/unbatch_to_individual_items.py
+++ b/rllib/connectors/module_to_env/unbatch_to_individual_items.py
@@ -8,10 +8,32 @@
 from ray.rllib.utils.annotations import override
 from ray.rllib.utils.spaces.space_utils import unbatch
 from ray.rllib.utils.typing import EpisodeType
+from ray.util.annotations import PublicAPI
 
 
+@PublicAPI(stability="alpha")
 class UnBatchToIndividualItems(ConnectorV2):
-    """Unbatches the given `data` back into the individual-batch-items format."""
+    """Unbatches the given `data` back into the individual-batch-items format.
+
+    Note: This is one of the default module-to-env ConnectorV2 pieces that
+    are added automatically by RLlib into every module-to-env connector pipeline,
+    unless `config.add_default_connectors_to_module_to_env_pipeline` is set to
+    False.
+
+    The default module-to-env connector pipeline is:
+    [
+        GetActions,
+        TensorToNumpy,
+        UnBatchToIndividualItems,
+        ModuleToAgentUnmapping,  # only in multi-agent setups!
+        RemoveSingleTsTimeRankFromBatch,
+
+        [0 or more user defined ConnectorV2 pieces],
+
+        NormalizeAndClipActions,
+        ListifyDataForVectorEnv,
+    ]
+    """
 
     @override(ConnectorV2)
     def __call__(