refactor(dataset): split LeRobotDataset into DatasetReader & DatasetWriter (+ API cleanup) (#3180)

* refactor(dataset): split reader and writer

* chore(dataset): remove proxys

* refactor(dataset): better reader & writer encapsulation

* refactor(datasets): clean API + reduce leaky implementations

* refactor(dataset): API cleaning for writer, reader and meta

* refactor(dataset): expose writer & reader + other minor improvements

* refactor(dataset): improve teardown routine

* refactor(dataset): add hf_dataset property at the facade level

* chore(dataset): add init for datasset module

* docs(dataset): add docstrings for public API of the dataset classes

* tests(dataset): add tests for new classes

* fix(dataset): remove circular dependecy

src/lerobot/rl/buffer.py

@@ -563,7 +563,7 @@ class ReplayBuffer:
         )
 
         # Start writing images if needed
-        lerobot_dataset.start_image_writer(num_processes=0, num_threads=3)
+        lerobot_dataset.writer.start_image_writer(num_processes=0, num_threads=3)
 
         # Convert transitions into episodes and frames
 
@@ -603,10 +603,10 @@ class ReplayBuffer:
                 lerobot_dataset.save_episode()
 
         # Save any remaining frames in the buffer
-        if lerobot_dataset.episode_buffer["size"] > 0:
+        if lerobot_dataset.has_pending_frames():
             lerobot_dataset.save_episode()
 
-        lerobot_dataset.stop_image_writer()
+        lerobot_dataset.writer.stop_image_writer()
         lerobot_dataset.finalize()
 
         return lerobot_dataset

src/lerobot/rl/gym_manipulator.py

@@ -752,8 +752,7 @@ def replay_trajectory(
         episodes=[cfg.dataset.replay_episode],
         download_videos=False,
     )
-    episode_frames = dataset.hf_dataset.filter(lambda x: x["episode_index"] == cfg.dataset.replay_episode)
-    actions = episode_frames.select_columns(ACTION)
+    actions = dataset.select_columns(ACTION)
 
     _, info = env.reset()