Remove deprecated DROID conversion scripts and update README with new pipeline usage examples, including auto-scan and quick mode options for trajectory processing.

Author: Your Name

examples/droid_h5/.gitignore

@@ -0,0 +1 @@
+results/

examples/droid_h5/README.md

@@ -45,10 +45,39 @@ The pipeline consists of three main steps:
 
 ### Complete Pipeline (Recommended)
 
-**The easiest way is to use the complete pipeline that handles everything:**
+**The easiest way is to use the complete pipeline with auto-scan:**
 
 ```bash
-# Complete end-to-end pipeline: Download → Convert → Process → Validate
+# Quick mode: Use pre-defined sample trajectories (fastest for testing)
+python droid_hdf5_pipeline.py \
+    --auto-scan --quick-mode \
+    --num-trajectories 3 \
+    --output-dir ./droid_hdf5_results \
+    --question "Is this trajectory successful?" \
+    --max-workers 2
+
+# Full scan: Automatically discover and select from all available trajectories  
+python droid_hdf5_pipeline.py \
+    --auto-scan \
+    --num-trajectories 10 \
+    --output-dir ./droid_hdf5_results \
+    --question "Is this trajectory successful?" \
+    --max-workers 4
+
+# Balanced selection (70% success, 30% failure) with reproducible results
+python droid_hdf5_pipeline.py \
+    --auto-scan --quick-mode \
+    --num-trajectories 20 \
+    --balance 0.7 \
+    --seed 42 \
+    --output-dir ./results \
+    --question "Did the robot complete the task successfully?"
+```
+
+**Legacy manual specification:**
+
+```bash
+# Manual trajectory specification
 python droid_hdf5_pipeline.py \
     --trajectories gs://gresearch/robotics/droid_raw/1.0.1/success/2023-07-21_16-18-07 \
                   gs://gresearch/robotics/droid_raw/1.0.1/failure/2023-07-21_16-27-21 \

examples/droid_h5/convert_droid_to_hdf5.py

@@ -0,0 +1,65 @@
+#!/usr/bin/env python3
+"""
+Create ground truth file from DROID metadata for validation.
+"""
+
+import json
+import os
+import glob
+from pathlib import Path
+
+def create_ground_truth_from_metadata(results_dir, output_file):
+    """
+    Create a manual ground truth file from DROID metadata files.
+    
+    Args:
+        results_dir: Directory containing DROID trajectories  
+        output_file: Output JSON file for ground truth labels
+    """
+    ground_truth = {}
+    
+    # Find all metadata files
+    metadata_files = glob.glob(os.path.join(results_dir, "droid_trajectories", "*", "metadata_*.json"))
+    
+    for metadata_file in metadata_files:
+        try:
+            with open(metadata_file, 'r') as f:
+                metadata = json.load(f)
+            
+            # Extract trajectory directory name
+            trajectory_dir = os.path.dirname(metadata_file)
+            trajectory_name = os.path.basename(trajectory_dir)
+            
+            # Create the path format used in VLM results
+            trajectory_path = f"./results/droid_trajectories/{trajectory_name}"
+            
+            # Extract success label
+            success = metadata.get("success", None)
+            if success is not None:
+                ground_truth[trajectory_path] = success
+                print(f"Added: {trajectory_name} -> {success}")
+                
+        except Exception as e:
+            print(f"Error processing {metadata_file}: {e}")
+            continue
+    
+    # Save ground truth file
+    with open(output_file, 'w') as f:
+        json.dump(ground_truth, f, indent=2)
+    
+    print(f"\nCreated ground truth file: {output_file}")
+    print(f"Total trajectories: {len(ground_truth)}")
+    
+    # Count success/failure
+    successful = sum(1 for v in ground_truth.values() if v)
+    failed = sum(1 for v in ground_truth.values() if not v)
+    print(f"Successful: {successful}")
+    print(f"Failed: {failed}")
+    
+    return ground_truth
+
+if __name__ == "__main__":
+    results_dir = "./results"
+    output_file = "./results/ground_truth.json"
+    
+    create_ground_truth_from_metadata(results_dir, output_file)