Merge branch 'feat/add_relative_action_pi_models' into feat/mirror

src/lerobot/scripts/lerobot_train.py

@@ -243,13 +243,35 @@ def train(cfg: TrainPipelineConfig, accelerator: Accelerator | None = None):
         peft_cli_overrides = dataclasses.asdict(cfg.peft)
         policy = policy.wrap_with_peft(peft_cli_overrides=peft_cli_overrides)
 
-    # Recompute action stats as delta if use_delta_actions is enabled
+    # Recompute action stats as delta if use_delta_actions is enabled.
+    # Must iterate the actual dataset (which returns action chunks via delta_timestamps)
+    # so stats capture the full range of chunk-level deltas, not just per-frame deltas.
     if getattr(cfg.policy, "use_delta_actions", False) and is_main_process:
-        logging.info("use_delta_actions is enabled — recomputing action stats as delta (action - state)")
+        logging.info("use_delta_actions is enabled — computing delta action stats from dataset chunks")
-        from lerobot.datasets.dataset_tools import recompute_stats
+        from lerobot.datasets.compute_stats import get_feature_stats
+        from lerobot.processor.delta_action_processor import to_delta_actions
 
-        exclude = getattr(cfg.policy, "delta_exclude_joints", [])
+        all_delta_actions = []
-        recompute_stats(dataset, skip_image_video=True, delta_action=True, delta_exclude_joints=exclude)
+        for i in range(len(dataset)):
+            item = dataset[i]
+            action = item["action"]
+            state = item["observation.state"]
+            # action may be (chunk_size, action_dim) or (action_dim,)
+            if action.ndim == 1:
+                action = action.unsqueeze(0)
+            mask = [True] * action.shape[-1]
+            delta = to_delta_actions(action.unsqueeze(0), state.unsqueeze(0), mask).squeeze(0)
+            all_delta_actions.append(delta.numpy())
+
+        import numpy as np
+
+        all_delta = np.concatenate(all_delta_actions, axis=0)
+        delta_stats = get_feature_stats(all_delta, axis=0, keepdims=all_delta.ndim == 1)
+        dataset.meta.stats["action"] = delta_stats
+        logging.info(
+            f"Delta action stats computed from {len(dataset)} samples: "
+            f"mean={np.abs(delta_stats['mean']).mean():.4f}, std={delta_stats['std'].mean():.4f}"
+        )
 
     # Wait for all processes to finish policy creation before continuing
     accelerator.wait_for_everyone()