diff --git a/docs/source/unitree_g1.mdx b/docs/source/unitree_g1.mdx
index a7a25c73b..69965a561 100644
--- a/docs/source/unitree_g1.mdx
+++ b/docs/source/unitree_g1.mdx
@@ -274,7 +274,8 @@ python src/lerobot/scripts/lerobot_train.py \
 Once trained, we recommend deploying policies using inference-time RTC:
 
 ```bash
-python examples/rtc/eval_with_real_robot.py \
+lerobot-rollout \
+  --strategy.type=base \
   --policy.path=your-username/your-repo-id \
   --policy.device=cuda \
   --robot.type=unitree_g1 \
diff --git a/src/lerobot/rollout/configs.py b/src/lerobot/rollout/configs.py
index 46f46eae1..d1527dc08 100644
--- a/src/lerobot/rollout/configs.py
+++ b/src/lerobot/rollout/configs.py
@@ -142,7 +142,9 @@ class DAggerStrategyConfig(RolloutStrategyConfig):
     windows, where each correction becomes its own episode.
     """
 
-    num_episodes: int = 10
+    # Number of correction episodes to collect (corrections-only mode).
+    # When None, falls back to ``--dataset.num_episodes``.
+    num_episodes: int | None = None
     record_autonomous: bool = False
     upload_every_n_episodes: int = 5
     # Target video file size in MB for episode rotation (record_autonomous
@@ -258,6 +260,19 @@ class RolloutConfig:
                     "--dataset.streaming_encoding=true --dataset.encoder_threads=2"
                 )
 
+        # DAgger: resolve num_episodes from dataset config when not explicitly set.
+        if isinstance(self.strategy, DAggerStrategyConfig) and self.strategy.num_episodes is None:
+            if self.dataset is not None:
+                self.strategy.num_episodes = self.dataset.num_episodes
+                logger.info(
+                    "DAgger num_episodes not set — using --dataset.num_episodes=%d",
+                    self.strategy.num_episodes,
+                )
+            else:
+                raise ValueError(
+                    "DAgger num_episodes must be set either via --strategy.num_episodes or --dataset.num_episodes"
+                )
+
         # --- Policy loading ---
         if self.robot is None:
             raise ValueError("--robot.type is required for rollout")
diff --git a/tests/policies/rtc/test_rtc_relative_actions.py b/tests/policies/rtc/test_rtc_relative_actions.py
index fa888ec05..14c115764 100644
--- a/tests/policies/rtc/test_rtc_relative_actions.py
+++ b/tests/policies/rtc/test_rtc_relative_actions.py
@@ -187,7 +187,7 @@ class TestRTCDenoiseWithRelativeLeftovers:
 
 
 class TestFullPipelineRelativeRTC:
-    """End-to-end test of the RTC + relative actions pipeline matching eval_with_real_robot.py flow."""
+    """End-to-end test of the RTC + relative actions pipeline matching lerobot-rollout flow."""
 
     def test_preprocessor_caches_state_for_postprocessor(self):
         """Preprocessor's relative step should cache state so postprocessor can convert back."""
@@ -240,7 +240,7 @@ class TestFullPipelineRelativeRTC:
         torch.testing.assert_close(recovered, actions, atol=1e-5, rtol=1e-5)
 
     def test_eval_loop_simulation(self):
-        """Simulate the eval_with_real_robot.py loop with relative actions.
+        """Simulate the lerobot-rollout loop with relative actions.
 
         Iteration 1: No leftovers → model generates relative actions → store for RTC
         Iteration 2: Use leftovers as RTC guidance → model generates new relative actions
@@ -401,12 +401,12 @@ class TestStateRebasingApproximation:
 
 
 def _detect_relative_actions(preprocessor) -> bool:
-    """Mirror of the helper in eval_with_real_robot.py for testing without importing it."""
+    """Mirror of the helper in lerobot-rollout for testing without importing it."""
     return any(isinstance(step, RelativeActionsProcessorStep) and step.enabled for step in preprocessor.steps)
 
 
 class TestDetectRelativeActions:
-    """Test the _detect_relative_actions helper logic used by eval_with_real_robot.py."""
+    """Test the _detect_relative_actions helper logic used by lerobot-rollout."""
 
     def test_detects_enabled_relative_step(self):
         class FakePipeline: