build obs with policy names

examples/openarms/evaluate_ee.py

@@ -319,8 +319,9 @@ def run_ee_inference_loop(
         ee_state = joints_to_ee(joint_state.copy())
         
         # 3. Build observation frame with EE state for policy input
-        # Use state names from dataset features to match training order
-        state_names = dataset.features.get("observation.state", {}).get("names", [])
+        # Use state names from policy's input features (from training) to match expected order
+        state_feature = policy.config.input_features.get("observation.state")
+        state_names = getattr(state_feature, "names", None) if state_feature else None
         
         if state_names:
             # Build state array using the exact names from training
@@ -518,12 +519,14 @@ def main():
     policy_config = PreTrainedConfig.from_pretrained(HF_MODEL_ID)
     policy_config.pretrained_path = HF_MODEL_ID
     
-    policy = make_policy(policy_config, ds_meta=dataset.meta)
+    # Create policy without new dataset meta (use pretrained config)
+    policy = make_policy(policy_config, ds_meta=None)
     
+    # Use pretrained stats from model, NOT from new evaluation dataset
+    # (evaluation dataset has different features than training dataset)
     preprocessor, postprocessor = make_pre_post_processors(
         policy_cfg=policy.config,
         pretrained_path=HF_MODEL_ID,
-        dataset_stats=dataset.meta.stats,
         preprocessor_overrides={
             "device_processor": {"device": str(policy.config.device)}
         },