droid/scripts/debug_cartesian_position.py

"""
Debug what cartesian_position represents in DROID data.

Compare:
1. cartesian_position from observation
2. FK from joint_position using different links (wrist, flange, gripper)
"""

import sys
from pathlib import Path
sys.path.append(str(Path(__file__).parent.parent))

import os
import numpy as np

# Import torch first
import torch
import mediapy as media

# Import TensorFlow and configure for CPU
import tensorflow as tf
tf.config.set_visible_devices([], 'GPU')
import tensorflow_datasets as tfds

import cv2
import datetime
import re
import pybullet as p

from utils.load_camera_calibration import CameraCalibrationLoader
from utils.franka_mesh_projection import FrankaMeshProjector


def load_cotracker():
    """Load CoTracker v3 model."""
    from cotracker.predictor import CoTrackerPredictor
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    model = CoTrackerPredictor(checkpoint='/mnt/kevin/vlm_models/cotracker/scaled_offline.pth')
    model = model.to(device)
    model.eval()
    return model, device


def find_closest_calibration(episode, uuid_list):
    """Find closest calibration by timestamp."""
    try:
        recording_path = episode['episode_metadata']['recording_folderpath'].numpy().decode('utf-8')
        match = re.search(r'/([A-Z]+)/success/(\d{4}-\d{2}-\d{2})/\w+_\w+_+\d+_(\d{2}):(\d{2}):(\d{2})_\d{4}/', recording_path)
        if not match:
            return None

        lab, date, hour, minute, second = match.groups()
        episode_time = datetime.datetime.strptime(f"{date} {hour}:{minute}:{second}", "%Y-%m-%d %H:%M:%S")

        matching_calibs = [uuid for uuid in uuid_list if uuid.startswith(f"{lab}+") and f"+{date}-" in uuid]
        if len(matching_calibs) == 0:
            return None

        best_uuid = None
        min_time_diff = float('inf')
        for calib_uuid in matching_calibs:
            parts = calib_uuid.split('+')
            if len(parts) >= 3:
                time_str = parts[2].replace('_cameras', '')
                match_time = re.search(r'(\d{2})h-(\d{2})m-(\d{2})s', time_str)
                if match_time:
                    calib_hour = int(match_time.group(1))
                    calib_min = int(match_time.group(2))
                    calib_sec = int(match_time.group(3))
                    calib_time = datetime.datetime.strptime(
                        f"{date} {calib_hour}:{calib_min}:{calib_sec}",
                        "%Y-%m-%d %H:%M:%S"
                    )
                    time_diff = abs((episode_time - calib_time).total_seconds())
                    if time_diff < min_time_diff:
                        min_time_diff = time_diff
                        best_uuid = calib_uuid
        return best_uuid
    except Exception as e:
        return None


def get_link_position_from_fk(projector, joint_positions, link_idx):
    """Get 3D position of a specific link using FK."""
    # Set joint positions
    for i in range(min(7, projector.num_joints)):
        p.resetJointState(projector.robot_id, i, joint_positions[i])

    # Get link state
    link_state = p.getLinkState(projector.robot_id, link_idx)
    link_pos = np.array(link_state[4])  # world position
    return link_pos


def main():
    output_dir = Path('/tmp/droid_debug_cartesian')
    output_dir.mkdir(parents=True, exist_ok=True)

    print("=" * 80)
    print("Debugging cartesian_position vs FK")
    print("=" * 80)

    # Initialize
    calib_dir = '/root/workspace/code/wmrl/Dual-Dynamics-Models/DROID-main/vision/u/wenlongh/datasets/droid_v4/cameras'
    calib_loader = CameraCalibrationLoader(calib_dir)
    projector = FrankaMeshProjector(use_gui=False)
    cotracker, device = load_cotracker()

    # Get UUID list
    calib_path = Path(calib_dir)
    calib_files = sorted(calib_path.glob("*_cameras.json"))
    uuid_list = [f.stem.replace('_cameras', '') for f in calib_files]

    # Load dataset
    droid_path = '/mnt/kevin/data/droid/droid/1.0.0'
    print("Loading DROID dataset...")
    builder = tfds.builder_from_directory(droid_path)
    dataset = builder.as_dataset(split='train')

    # Find first valid episode
    episode_found = None
    uuid_found = None

    for episode_idx, episode in enumerate(dataset):
        uuid = find_closest_calibration(episode, uuid_list)
        if uuid is None:
            continue
        if not calib_loader.has_refined_extrinsics(uuid):
            continue

        episode_found = episode
        uuid_found = uuid
        print(f"\nUsing episode {episode_idx}, UUID: {uuid}")
        break

    if episode_found is None:
        print("No valid episode found!")
        return

    # Get calibration
    calib = calib_loader.load_calibration(uuid_found)
    available_serials = [k for k in calib.keys() if k not in ['uuid', 'scene_path', 'optimization_summary']]
    camera_serial = available_serials[0]  # exterior
    cam_data = calib[camera_serial]

    K = np.array(cam_data['measured_intrinsics'])
    E = np.array(cam_data['refined_extrinsics'])

    # Collect frames
    frames = []
    cart_positions = []
    joint_positions_list = []
    max_frames = 16

    for step_idx, step in enumerate(episode_found['steps']):
        if step_idx >= max_frames:
            break

        cart_pos = step['observation']['cartesian_position'].numpy()
        joint_pos = step['observation']['joint_position'].numpy()
        img = step['observation']['exterior_image_1_left'].numpy()

        cart_positions.append(cart_pos)
        joint_positions_list.append(joint_pos)
        frames.append(img)

    img_h, img_w = frames[0].shape[:2]

    # Compare cartesian_position vs FK for different links
    print("\n" + "=" * 80)
    print("Comparing cartesian_position vs FK positions")
    print("=" * 80)

    cart_pos_0 = cart_positions[0]
    joint_pos_0 = joint_positions_list[0]

    print(f"\ncartesian_position: {cart_pos_0[:3]}")
    print(f"joint_position: {joint_pos_0}")

    # Test different links
    link_names = {
        5: "Link 5 (wrist)",
        7: "Link 7 (flange)",
        8: "Link 8 (panda_hand)",
        9: "Link 9 (panda_leftfinger)",
        10: "Link 10 (panda_rightfinger)",
    }

    fk_positions = {}
    for link_idx, link_name in link_names.items():
        fk_pos = get_link_position_from_fk(projector, joint_pos_0, link_idx)
        fk_positions[link_idx] = fk_pos
        diff = np.linalg.norm(fk_pos - cart_pos_0[:3])
        print(f"\n{link_name:25s}: {fk_pos}")
        print(f"  Distance from cartesian_position: {diff:.4f}")

    # Find closest link
    closest_link = min(fk_positions.items(), key=lambda x: np.linalg.norm(x[1] - cart_pos_0[:3]))
    print(f"\nClosest match: {link_names[closest_link[0]]}")

    # Now visualize tracking each link
    print("\n" + "=" * 80)
    print("Generating visualization videos")
    print("=" * 80)

    for link_idx, link_name in link_names.items():
        print(f"\nProcessing {link_name}...")

        # Project this link for all frames
        all_projections = []
        for frame_idx, joint_pos in enumerate(joint_positions_list):
            link_pos = get_link_position_from_fk(projector, joint_pos, link_idx)
            link_pos_reshaped = link_pos.reshape(1, 3)
            proj_2d, proj_vis = projector._project_3d_to_2d(link_pos_reshaped, K, E, img_h=img_h, img_w=img_w)
            all_projections.append((proj_2d[0], proj_vis[0]))

        # Get initial projection for CoTracker
        init_2d, init_vis = all_projections[0]
        if not init_vis:
            print(f"  Skipping {link_name} - not visible")
            continue

        # Run CoTracker
        video_np = np.array(frames)
        video_np = video_np.transpose(0, 3, 1, 2)
        video_tensor = torch.from_numpy(video_np).float() / 255.0
        video_tensor = video_tensor.unsqueeze(0).to(device)

        queries = np.zeros((1, 3))
        queries[0, 0] = 0
        queries[0, 1] = init_2d[0]
        queries[0, 2] = init_2d[1]
        queries_tensor = torch.from_numpy(queries).float().unsqueeze(0).to(device)

        with torch.no_grad():
            pred_tracks, pred_visibility = cotracker(
                video_tensor,
                queries=queries_tensor,
                backward_tracking=False
            )

        tracks = pred_tracks[0].cpu().numpy()
        visibility = pred_visibility[0].cpu().numpy()

        # Visualize
        video_frames = []
        for frame_idx, frame in enumerate(frames):
            viz = frame.copy()

            # Draw GT projection for this frame (blue)
            gt_2d, gt_vis = all_projections[frame_idx]
            if gt_vis:
                gt_pt = tuple(gt_2d.astype(int))
                cv2.circle(viz, gt_pt, 5, (255, 0, 0), 2)  # Blue circle

            # Draw tracked point (green)
            if visibility[frame_idx, 0]:
                track_pt = tuple(tracks[frame_idx, 0].astype(int))
                cv2.circle(viz, track_pt, 3, (0, 255, 0), -1)  # Green filled

                # Draw trajectory
                if frame_idx > 0:
                    for t in range(max(0, frame_idx-10), frame_idx):
                        if visibility[t, 0] and visibility[t+1, 0]:
                            pt1 = tuple(tracks[t, 0].astype(int))
                            pt2 = tuple(tracks[t+1, 0].astype(int))
                            cv2.line(viz, pt1, pt2, (0, 255, 0), 1)

            # Add text
            cv2.putText(viz, link_name, (10, 25), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (255, 255, 255), 2)
            cv2.putText(viz, f"Frame {frame_idx}/{len(frames)}", (10, 50), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1)
            cv2.putText(viz, f"Blue=GT, Green=Tracked", (10, 70), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (255, 255, 0), 1)

            video_frames.append(viz)

        # Save video
        video_name = f"{link_name.replace(' ', '_').replace('(', '').replace(')', '')}.mp4"
        video_path = output_dir / video_name
        media.write_video(str(video_path), video_frames, fps=10)
        print(f"  Saved: {video_path}")

    print("\n" + "=" * 80)
    print(f"Complete! Videos saved to: {output_dir}")
    print("=" * 80)


if __name__ == "__main__":
    main()