Merge branch 'main' into codex/language-recipes

src/lerobot/policies/gaussian_actor/configuration_gaussian_actor.py

@@ -37,13 +37,19 @@ def is_image_feature(key: str) -> bool:
 @dataclass
 class ConcurrencyConfig:
     """Configuration for the concurrency of the actor and learner.
+
     Possible values are:
     - "threads": Use threads for the actor and learner.
     - "processes": Use processes for the actor and learner.
+
+    ``multiprocessing_context`` selects the process-wide start method when
+    processes are used. Set it to ``None`` to preserve Python's default or a
+    method already selected by the embedding application.
     """
 
     actor: str = "threads"
     learner: str = "threads"
+    multiprocessing_context: str | None = "spawn"
 
 
 @dataclass

src/lerobot/processor/delta_action_processor.py

@@ -132,10 +132,20 @@ class MapDeltaActionToRobotActionStep(RobotActionProcessorStep):
     def transform_features(
         self, features: dict[PipelineFeatureType, dict[str, PolicyFeature]]
     ) -> dict[PipelineFeatureType, dict[str, PolicyFeature]]:
-        for axis in ["x", "y", "z", "gripper"]:
+        for axis in ["x", "y", "z"]:
             features[PipelineFeatureType.ACTION].pop(f"delta_{axis}", None)
+        features[PipelineFeatureType.ACTION].pop("gripper", None)
 
-        for feat in ["enabled", "target_x", "target_y", "target_z", "target_wx", "target_wy", "target_wz"]:
+        for feat in [
+            "enabled",
+            "target_x",
+            "target_y",
+            "target_z",
+            "target_wx",
+            "target_wy",
+            "target_wz",
+            "gripper_vel",
+        ]:
             features[PipelineFeatureType.ACTION][f"{feat}"] = PolicyFeature(
                 type=FeatureType.ACTION, shape=(1,)
             )

src/lerobot/rl/actor.py

@@ -91,7 +91,7 @@ from lerobot.robots import so_follower  # noqa: F401
 from lerobot.teleoperators import gamepad, so_leader  # noqa: F401
 from lerobot.teleoperators.utils import TeleopEvents
 from lerobot.utils.device_utils import get_safe_torch_device
-from lerobot.utils.process import ProcessSignalHandler
+from lerobot.utils.process import ProcessSignalHandler, ensure_multiprocessing_start_method
 from lerobot.utils.random_utils import set_seed
 from lerobot.utils.robot_utils import precise_sleep
 from lerobot.utils.transition import (
@@ -124,9 +124,7 @@ def actor_cli(cfg: TrainRLServerPipelineConfig):
     cfg.validate()
     display_pid = False
     if not use_threads(cfg):
-        import torch.multiprocessing as mp
-
-        mp.set_start_method("spawn")
+        ensure_multiprocessing_start_method(cfg.policy.concurrency.multiprocessing_context)
         display_pid = True
 
     # Create logs directory to ensure it exists

src/lerobot/rl/learner.py

@@ -102,7 +102,7 @@ from lerobot.utils.constants import (
 )
 from lerobot.utils.device_utils import get_safe_torch_device
 from lerobot.utils.io_utils import load_json, write_json
-from lerobot.utils.process import ProcessSignalHandler
+from lerobot.utils.process import ProcessSignalHandler, ensure_multiprocessing_start_method
 from lerobot.utils.random_utils import set_seed
 from lerobot.utils.utils import (
     format_big_number,
@@ -123,9 +123,7 @@ def train_cli(cfg: TrainRLServerPipelineConfig):
     # Fail fast with a friendly error if the optional ``hilserl`` extra is missing.
     require_package("grpcio", extra="hilserl", import_name="grpc")
     if not use_threads(cfg):
-        import torch.multiprocessing as mp
-
-        mp.set_start_method("spawn")
+        ensure_multiprocessing_start_method(cfg.policy.concurrency.multiprocessing_context)
 
     # Use the job_name from the config
     train(

src/lerobot/utils/process.py

@@ -16,11 +16,39 @@
 # limitations under the License.
 
 import logging
+import multiprocessing
 import os
 import signal
 import sys
 
 
+def ensure_multiprocessing_start_method(start_method: str | None) -> None:
+    """Set a multiprocessing start method once, or verify the existing method matches.
+
+    Passing ``None`` leaves Python's process-wide default untouched. This is useful
+    when LeRobot is embedded in an application that owns multiprocessing setup.
+    """
+    if start_method is None:
+        return
+
+    available_methods = multiprocessing.get_all_start_methods()
+    if start_method not in available_methods:
+        raise ValueError(
+            f"Multiprocessing start method must be one of {available_methods} on this platform, "
+            f"got {start_method!r}."
+        )
+
+    current_method = multiprocessing.get_start_method(allow_none=True)
+    if current_method is None:
+        multiprocessing.set_start_method(start_method)
+    elif current_method != start_method:
+        raise RuntimeError(
+            f"Multiprocessing start method is already {current_method!r}; cannot change it to "
+            f"{start_method!r}. Set the configured multiprocessing context to null to keep the "
+            "application's existing method, or launch LeRobot in a fresh process."
+        )
+
+
 class ProcessSignalHandler:
     """Utility class to attach graceful shutdown signal handlers.
 

tests/policies/test_gaussian_actor_config.py

@@ -113,6 +113,7 @@ def test_gaussian_actor_config_default_initialization():
     # Concurrency configuration
     assert config.concurrency.actor == "threads"
     assert config.concurrency.learner == "threads"
+    assert config.concurrency.multiprocessing_context == "spawn"
 
     assert isinstance(config.actor_network_kwargs, ActorNetworkConfig)
     assert isinstance(config.policy_kwargs, PolicyConfig)
@@ -152,6 +153,7 @@ def test_concurrency_config():
     config = ConcurrencyConfig()
     assert config.actor == "threads"
     assert config.learner == "threads"
+    assert config.multiprocessing_context == "spawn"
 
 
 def test_gaussian_actor_config_custom_initialization():

tests/utils/test_collate.py

@@ -9,7 +9,7 @@ import torch  # noqa: E402
 from lerobot.utils.collate import lerobot_collate_fn  # noqa: E402
 
 
-def test_lerobot_collate_preserves_messages_and_drops_raw_language():
+def test_lerobot_collate_preserves_messages_and_raw_language():
     batch = [
         {
             "index": torch.tensor(0),
@@ -17,14 +17,14 @@ def test_lerobot_collate_preserves_messages_and_drops_raw_language():
             "message_streams": ["low_level"],
             "target_message_indices": [0],
             "language_persistent": [{"content": "raw"}],
-            "language_events": [],
+            "language_events": [{"content": "event a"}],
         },
         {
             "index": torch.tensor(1),
             "messages": [{"role": "assistant", "content": "b"}],
             "message_streams": ["low_level"],
             "target_message_indices": [0],
-            "language_persistent": [{"content": "raw"}],
+            "language_persistent": [{"content": "raw b"}],
             "language_events": [],
         },
     ]
@@ -36,8 +36,8 @@ def test_lerobot_collate_preserves_messages_and_drops_raw_language():
     assert out["messages"][1][0]["content"] == "b"
     assert out["message_streams"] == [["low_level"], ["low_level"]]
     assert out["target_message_indices"] == [[0], [0]]
-    assert "language_persistent" not in out
-    assert "language_events" not in out
+    assert out["language_persistent"] == [[{"content": "raw"}], [{"content": "raw b"}]]
+    assert out["language_events"] == [[{"content": "event a"}], []]
 
 
 def test_lerobot_collate_passes_through_standard_batch():