Merge remote-tracking branch 'origin/main' into user/khalil-meftah/2026-02-16-rl-stack-refactor

# Conflicts:
#	uv.lock

.github/workflows/stale.yml

@@ -19,8 +19,8 @@ on:
   workflow_dispatch:
 
   # Runs at 02:00
-  schedule:
+  # schedule:
-    - cron: "0 2 * * *"
+  #   - cron: "0 2 * * *"
 
 env:
   CLOSE_ISSUE_MESSAGE: >

pyproject.toml

@@ -59,8 +59,8 @@ keywords = ["lerobot", "huggingface", "robotics",  "machine learning", "artifici
 
 dependencies = [
     # Core ML
-    "torch>=2.7,<2.11.0",
+    "torch>=2.7,<2.13.0",
-    "torchvision>=0.22.0,<0.26.0",
+    "torchvision>=0.22.0,<0.28.0",
     "numpy>=2.0.0,<2.3.0", # NOTE: Explicitly listing numpy helps the resolver converge faster. Upper bound imposed by opencv-python-headless.
     "opencv-python-headless>=4.9.0,<4.14.0",
     "Pillow>=10.0.0,<13.0.0",
@@ -99,7 +99,7 @@ dataset = [
     "pandas>=2.0.0,<3.0.0", # NOTE: Transitive dependency of datasets
     "pyarrow>=21.0.0,<30.0.0", # NOTE: Transitive dependency of datasets
     "lerobot[av-dep]",
-    "torchcodec>=0.3.0,<0.11.0; sys_platform != 'win32' and (sys_platform != 'linux' or (platform_machine != 'aarch64' and platform_machine != 'arm64' and platform_machine != 'armv7l')) and (sys_platform != 'darwin' or platform_machine != 'x86_64')", # NOTE: Windows support starts at version 0.7 (needs torch==2.8), ffmpeg>=8 support starts at version 0.8.1 (needs torch==2.9), system-wide ffmpeg support starts at version 0.10 (needs torch==2.10).
+    "torchcodec>=0.3.0,<0.13.0; sys_platform != 'win32' and (sys_platform != 'linux' or (platform_machine != 'aarch64' and platform_machine != 'arm64' and platform_machine != 'armv7l')) and (sys_platform != 'darwin' or platform_machine != 'x86_64')", # NOTE: Windows support starts at version 0.7 (needs torch==2.8), ffmpeg>=8 support starts at version 0.8.1 (needs torch==2.9), system-wide ffmpeg support starts at version 0.10 (needs torch==2.10), 0.11 needs torch==2.11, 0.12 needs torch==2.12.
     "jsonlines>=4.0.0,<5.0.0",
 ]
 training = [
@@ -195,7 +195,7 @@ groot = [
 sarm = ["lerobot[transformers-dep]", "pydantic>=2.0.0,<3.0.0", "faker>=33.0.0,<35.0.0", "lerobot[matplotlib-dep]", "lerobot[qwen-vl-utils-dep]"]
 xvla = ["lerobot[transformers-dep]"]
 eo1 = ["lerobot[transformers-dep]", "lerobot[qwen-vl-utils-dep]"]
-hilserl = ["lerobot[transformers-dep]", "gym-hil>=0.1.13,<0.2.0", "lerobot[grpcio-dep]", "lerobot[placo-dep]"]
+hilserl = ["lerobot[transformers-dep]", "lerobot[dataset]", "gym-hil>=0.1.13,<0.2.0", "lerobot[grpcio-dep]", "lerobot[placo-dep]"]
 
 # Features
 async = ["lerobot[grpcio-dep]", "lerobot[matplotlib-dep]"]

src/lerobot/configs/train.py

@@ -256,7 +256,9 @@ class TrainPipelineConfig(HubMixin):
                 ) from e
 
         cli_args = kwargs.pop("cli_args", [])
-        if config_file is not None:
+        # Legacy RA-BC migration only applies to framework-saved checkpoints (always JSON).
+        # Hand-written YAML/TOML configs are expected to use the current sample_weighting schema.
+        if config_file is not None and config_file.endswith(".json"):
             with open(config_file) as f:
                 config = json.load(f)
             migrated_config = _migrate_legacy_rabc_fields(config)

src/lerobot/datasets/video_utils.py

@@ -282,7 +282,11 @@ class VideoDecoderCache:
         with self._lock:
             if video_path not in self._cache:
                 file_handle = fsspec.open(video_path).__enter__()
+                try:
                     decoder = VideoDecoder(file_handle, seek_mode="approximate")
+                except Exception:
+                    file_handle.close()
+                    raise
                 self._cache[video_path] = (decoder, file_handle)
 
             return self._cache[video_path][0]

src/lerobot/policies/wall_x/qwen_model/qwen2_5_vl_moe.py

@@ -939,7 +939,7 @@ class Qwen2_5_VLFlashAttention2(Qwen2_5_VLAttention):
         input_dtype = query_states.dtype
         if input_dtype == torch.float32:
             if torch.is_autocast_enabled():
-                target_dtype = torch.get_autocast_gpu_dtype()
+                target_dtype = torch.get_autocast_dtype(query_states.device.type)
             # Handle the case where the model is quantized
             elif hasattr(self.config, "_pre_quantization_dtype"):
                 target_dtype = self.config._pre_quantization_dtype

src/lerobot/policies/xvla/modeling_florence2.py

@@ -985,7 +985,7 @@ class Florence2FlashAttention2(Florence2Attention):
         input_dtype = query_states.dtype
         if input_dtype == torch.float32:
             if torch.is_autocast_enabled():
-                target_dtype = torch.get_autocast_gpu_dtype()
+                target_dtype = torch.get_autocast_dtype(query_states.device.type)
             # Handle the case where the model is quantized
             elif hasattr(self.config, "_pre_quantization_dtype"):
                 target_dtype = self.config._pre_quantization_dtype

src/lerobot/rl/actor.py

@@ -52,27 +52,15 @@ import time
 from collections.abc import Generator
 from functools import lru_cache
 from queue import Empty
-from typing import Any
+from typing import TYPE_CHECKING, Any
 
-from lerobot.utils.import_utils import require_package
+from lerobot.utils.import_utils import _grpc_available, require_package
 
-# Fail fast with a friendly error if the optional ``hilserl`` extra is missing.
+if TYPE_CHECKING or _grpc_available:
-require_package("grpcio", extra="hilserl", import_name="grpc")
+    import grpc
 
-import grpc  # noqa: E402
+    from lerobot.transport import services_pb2, services_pb2_grpc
-import torch  # noqa: E402
+    from lerobot.transport.utils import (
-from torch import nn  # noqa: E402
-from torch.multiprocessing import Queue  # noqa: E402
-
-from lerobot.cameras import opencv  # noqa: F401, E402
-from lerobot.configs import parser  # noqa: E402
-from lerobot.policies import make_policy, make_pre_post_processors  # noqa: E402
-from lerobot.processor import TransitionKey  # noqa: E402
-from lerobot.robots import so_follower  # noqa: F401, E402
-from lerobot.teleoperators import gamepad, so_leader  # noqa: F401, E402
-from lerobot.teleoperators.utils import TeleopEvents  # noqa: E402
-from lerobot.transport import services_pb2, services_pb2_grpc  # noqa: E402
-from lerobot.transport.utils import (  # noqa: E402
         bytes_to_state_dict,
         grpc_channel_options,
         python_object_to_bytes,
@@ -80,35 +68,59 @@ from lerobot.transport.utils import (  # noqa: E402
         send_bytes_in_chunks,
         transitions_to_bytes,
     )
-from lerobot.utils.device_utils import get_safe_torch_device  # noqa: E402
+else:
-from lerobot.utils.process import ProcessSignalHandler  # noqa: E402
+    grpc = None
-from lerobot.utils.random_utils import set_seed  # noqa: E402
+    services_pb2 = None
-from lerobot.utils.robot_utils import precise_sleep  # noqa: E402
+    services_pb2_grpc = None
-from lerobot.utils.transition import (  # noqa: E402
+    bytes_to_state_dict = None
+    grpc_channel_options = None
+    python_object_to_bytes = None
+    receive_bytes_in_chunks = None
+    send_bytes_in_chunks = None
+    transitions_to_bytes = None
+
+import torch
+from torch import nn
+from torch.multiprocessing import Queue
+
+from lerobot.cameras import opencv  # noqa: F401
+from lerobot.configs import parser
+from lerobot.policies import make_policy, make_pre_post_processors
+from lerobot.processor import TransitionKey
+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.random_utils import set_seed
+from lerobot.utils.robot_utils import precise_sleep
+from lerobot.utils.transition import (
     Transition,
     move_transition_to_device,
 )
-from lerobot.utils.utils import (  # noqa: E402
+from lerobot.utils.utils import (
     TimerManager,
     init_logging,
 )
 
-from .algorithms.base import RLAlgorithm  # noqa: E402
+from .algorithms.base import RLAlgorithm
-from .algorithms.factory import make_algorithm  # noqa: E402
+from .algorithms.factory import make_algorithm
-from .gym_manipulator import (  # noqa: E402
+from .gym_manipulator import (
     make_processors,
     make_robot_env,
     reset_and_build_transition,
     step_env_and_process_transition,
 )
-from .queue import get_last_item_from_queue  # noqa: E402
+from .queue import get_last_item_from_queue
-from .train_rl import TrainRLServerPipelineConfig  # noqa: E402
+from .train_rl import TrainRLServerPipelineConfig
 
 # Main entry point
 
 
 @parser.wrap()
 def actor_cli(cfg: TrainRLServerPipelineConfig):
+    # Fail fast with a friendly error if the optional ``hilserl`` extra is missing.
+    require_package("grpcio", extra="hilserl", import_name="grpc")
     cfg.validate()
     display_pid = False
     if not use_threads(cfg):
@@ -421,7 +433,7 @@ def act_with_policy(
 
 
 def establish_learner_connection(
-    stub: services_pb2_grpc.LearnerServiceStub,
+    stub: "services_pb2_grpc.LearnerServiceStub",
     shutdown_event: Any,  # Event
     attempts: int = 30,
 ) -> bool:
@@ -454,7 +466,7 @@ def establish_learner_connection(
 def learner_service_client(
     host: str = "127.0.0.1",
     port: int = 50051,
-) -> tuple[services_pb2_grpc.LearnerServiceStub, grpc.Channel]:
+) -> "tuple[services_pb2_grpc.LearnerServiceStub, grpc.Channel]":
     """Return a client for the learner service.
 
     GRPC uses HTTP/2, which is a binary protocol and multiplexes requests over a single connection.
@@ -477,8 +489,8 @@ def receive_policy(
     cfg: TrainRLServerPipelineConfig,
     parameters_queue: Queue,
     shutdown_event: Any,  # Event
-    learner_client: services_pb2_grpc.LearnerServiceStub | None = None,
+    learner_client: "services_pb2_grpc.LearnerServiceStub | None" = None,
-    grpc_channel: grpc.Channel | None = None,
+    grpc_channel: "grpc.Channel | None" = None,
 ) -> None:
     """Receive parameters from the learner.
 
@@ -531,8 +543,8 @@ def send_transitions(
     cfg: TrainRLServerPipelineConfig,
     transitions_queue: Queue,
     shutdown_event: Any,  # Event
-    learner_client: services_pb2_grpc.LearnerServiceStub | None = None,
+    learner_client: "services_pb2_grpc.LearnerServiceStub | None" = None,
-    grpc_channel: grpc.Channel | None = None,
+    grpc_channel: "grpc.Channel | None" = None,
 ) -> None:
     """Send transitions to the learner.
 
@@ -587,8 +599,8 @@ def send_interactions(
     cfg: TrainRLServerPipelineConfig,
     interactions_queue: Queue,
     shutdown_event: Any,  # Event
-    learner_client: services_pb2_grpc.LearnerServiceStub | None = None,
+    learner_client: "services_pb2_grpc.LearnerServiceStub | None" = None,
-    grpc_channel: grpc.Channel | None = None,
+    grpc_channel: "grpc.Channel | None" = None,
 ) -> None:
     """Send interactions to the learner.
 
@@ -646,7 +658,7 @@ def transitions_stream(
     shutdown_event: Any,  # Event
     transitions_queue: Queue,
     timeout: float,
-) -> Generator[Any, None, services_pb2.Empty]:
+) -> "Generator[Any, None, services_pb2.Empty]":
     while not shutdown_event.is_set():
         try:
             message = transitions_queue.get(block=True, timeout=timeout)
@@ -665,7 +677,7 @@ def interactions_stream(
     shutdown_event: Any,  # Event
     interactions_queue: Queue,
     timeout: float,
-) -> Generator[Any, None, services_pb2.Empty]:
+) -> "Generator[Any, None, services_pb2.Empty]":
     while not shutdown_event.is_set():
         try:
             message = interactions_queue.get(block=True, timeout=timeout)

src/lerobot/rl/learner.py

@@ -51,44 +51,47 @@ import time
 from concurrent.futures import ThreadPoolExecutor
 from pathlib import Path
 from pprint import pformat
-from typing import Any
+from typing import TYPE_CHECKING, Any
 
-from lerobot.utils.import_utils import require_package
+from lerobot.utils.import_utils import _grpc_available, require_package
 
-# Fail fast with a friendly error if the optional ``hilserl`` extra is missing.
+if TYPE_CHECKING or _grpc_available:
-require_package("grpcio", extra="hilserl", import_name="grpc")
+    import grpc
 
-import grpc  # noqa: E402
+    from lerobot.transport import services_pb2_grpc
-import torch  # noqa: E402
+else:
-from huggingface_hub.constants import SAFETENSORS_SINGLE_FILE  # noqa: E402
+    grpc = None
-from safetensors.torch import load_file as load_safetensors  # noqa: E402
+    services_pb2_grpc = None
-from termcolor import colored  # noqa: E402
-from torch import nn  # noqa: E402
-from torch.multiprocessing import Queue  # noqa: E402
-from torch.optim.optimizer import Optimizer  # noqa: E402
 
-from lerobot.cameras import opencv  # noqa: F401, E402
+import torch
-from lerobot.common.train_utils import (  # noqa: E402
+from huggingface_hub.constants import SAFETENSORS_SINGLE_FILE
+from safetensors.torch import load_file as load_safetensors
+from termcolor import colored
+from torch import nn
+from torch.multiprocessing import Queue
+from torch.optim.optimizer import Optimizer
+
+from lerobot.cameras import opencv  # noqa: F401
+from lerobot.common.train_utils import (
     get_step_checkpoint_dir,
     load_training_state as utils_load_training_state,
     save_checkpoint,
     update_last_checkpoint,
 )
-from lerobot.common.wandb_utils import WandBLogger  # noqa: E402
+from lerobot.common.wandb_utils import WandBLogger
-from lerobot.configs import parser  # noqa: E402
+from lerobot.configs import parser
-from lerobot.datasets import LeRobotDataset, make_dataset  # noqa: E402
+from lerobot.datasets import LeRobotDataset, make_dataset
-from lerobot.policies import make_policy, make_pre_post_processors  # noqa: E402
+from lerobot.policies import make_policy, make_pre_post_processors
-from lerobot.robots import so_follower  # noqa: F401, E402
+from lerobot.robots import so_follower  # noqa: F401
-from lerobot.teleoperators import gamepad, so_leader  # noqa: F401, E402
+from lerobot.teleoperators import gamepad, so_leader  # noqa: F401
-from lerobot.teleoperators.utils import TeleopEvents  # noqa: E402
+from lerobot.teleoperators.utils import TeleopEvents
-from lerobot.transport import services_pb2_grpc  # noqa: E402
+from lerobot.transport.utils import (
-from lerobot.transport.utils import (  # noqa: E402
     MAX_MESSAGE_SIZE,
     bytes_to_python_object,
     bytes_to_transitions,
     state_to_bytes,
 )
-from lerobot.utils.constants import (  # noqa: E402
+from lerobot.utils.constants import (
     ACTION,
     ALGORITHM_DIR,
     CHECKPOINTS_DIR,
@@ -97,26 +100,28 @@ from lerobot.utils.constants import (  # noqa: E402
     TRAINING_STATE_DIR,
     TRAINING_STEP,
 )
-from lerobot.utils.device_utils import get_safe_torch_device  # noqa: E402
+from lerobot.utils.device_utils import get_safe_torch_device
-from lerobot.utils.io_utils import load_json, write_json  # noqa: E402
+from lerobot.utils.io_utils import load_json, write_json
-from lerobot.utils.process import ProcessSignalHandler  # noqa: E402
+from lerobot.utils.process import ProcessSignalHandler
-from lerobot.utils.random_utils import set_seed  # noqa: E402
+from lerobot.utils.random_utils import set_seed
-from lerobot.utils.utils import (  # noqa: E402
+from lerobot.utils.utils import (
     format_big_number,
     init_logging,
 )
 
-from .algorithms.base import RLAlgorithm  # noqa: E402
+from .algorithms.base import RLAlgorithm
-from .algorithms.factory import make_algorithm  # noqa: E402
+from .algorithms.factory import make_algorithm
-from .buffer import ReplayBuffer  # noqa: E402
+from .buffer import ReplayBuffer
-from .data_sources import OnlineOfflineMixer  # noqa: E402
+from .data_sources import OnlineOfflineMixer
-from .learner_service import MAX_WORKERS, SHUTDOWN_TIMEOUT, LearnerService  # noqa: E402
+from .learner_service import MAX_WORKERS, SHUTDOWN_TIMEOUT, LearnerService
-from .train_rl import TrainRLServerPipelineConfig  # noqa: E402
+from .train_rl import TrainRLServerPipelineConfig
-from .trainer import RLTrainer  # noqa: E402
+from .trainer import RLTrainer
 
 
 @parser.wrap()
 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
 

src/lerobot/rl/learner_service.py

@@ -18,22 +18,32 @@
 import logging
 import time
 from multiprocessing import Event, Queue
+from typing import TYPE_CHECKING
 
-from lerobot.utils.import_utils import require_package
+from lerobot.utils.import_utils import _grpc_available
 
-# Fail fast with a friendly error if the optional ``hilserl`` extra is missing.
+from .queue import get_last_item_from_queue
-require_package("grpcio", extra="hilserl", import_name="grpc")
 
-from lerobot.transport import services_pb2, services_pb2_grpc  # noqa: E402
+if TYPE_CHECKING or _grpc_available:
-from lerobot.transport.utils import receive_bytes_in_chunks, send_bytes_in_chunks  # noqa: E402
+    import grpc
 
-from .queue import get_last_item_from_queue  # noqa: E402
+    from lerobot.transport import services_pb2, services_pb2_grpc
+    from lerobot.transport.utils import receive_bytes_in_chunks, send_bytes_in_chunks
+
+    _ServicerBase = services_pb2_grpc.LearnerServiceServicer
+else:
+    grpc = None
+    services_pb2 = None
+    services_pb2_grpc = None
+    receive_bytes_in_chunks = None
+    send_bytes_in_chunks = None
+    _ServicerBase = object
 
 MAX_WORKERS = 3  # Stream parameters, send transitions and interactions
 SHUTDOWN_TIMEOUT = 10
 
 
-class LearnerService(services_pb2_grpc.LearnerServiceServicer):
+class LearnerService(_ServicerBase):
     """
     Implementation of the LearnerService gRPC service
     This service is used to send parameters to the Actor and receive transitions and interactions from the Actor
@@ -56,7 +66,9 @@ class LearnerService(services_pb2_grpc.LearnerServiceServicer):
         self.interaction_message_queue = interaction_message_queue
         self.queue_get_timeout = queue_get_timeout
 
-    def StreamParameters(self, request, context):  # noqa: N802
+    def StreamParameters(  # noqa: N802
+        self, request: "services_pb2.Empty", context: "grpc.ServicerContext"
+    ):
         # TODO: authorize the request
         logging.info("[LEARNER] Received request to stream parameters from the Actor")
 
@@ -91,7 +103,7 @@ class LearnerService(services_pb2_grpc.LearnerServiceServicer):
         logging.info("[LEARNER] Stream parameters finished")
         return services_pb2.Empty()
 
-    def SendTransitions(self, request_iterator, _context):  # noqa: N802
+    def SendTransitions(self, request_iterator, _context: "grpc.ServicerContext"):  # noqa: N802
         # TODO: authorize the request
         logging.info("[LEARNER] Received request to receive transitions from the Actor")
 
@@ -105,7 +117,7 @@ class LearnerService(services_pb2_grpc.LearnerServiceServicer):
         logging.debug("[LEARNER] Finished receiving transitions")
         return services_pb2.Empty()
 
-    def SendInteractions(self, request_iterator, _context):  # noqa: N802
+    def SendInteractions(self, request_iterator, _context: "grpc.ServicerContext"):  # noqa: N802
         # TODO: authorize the request
         logging.info("[LEARNER] Received request to receive interactions from the Actor")
 
@@ -119,5 +131,5 @@ class LearnerService(services_pb2_grpc.LearnerServiceServicer):
         logging.debug("[LEARNER] Finished receiving interactions")
         return services_pb2.Empty()
 
-    def Ready(self, request, context):  # noqa: N802
+    def Ready(self, request: "services_pb2.Empty", context: "grpc.ServicerContext"):  # noqa: N802
         return services_pb2.Empty()

src/lerobot/robots/lekiwi/config_lekiwi.py

@@ -46,7 +46,7 @@ class LeKiwiConfig(RobotConfig):
     cameras: dict[str, CameraConfig] = field(default_factory=lekiwi_cameras_config)
 
     # Set to `True` for backward compatibility with previous policies/dataset
-    use_degrees: bool = False
+    use_degrees: bool = True
 
 
 @dataclass