droid/scripts/test_rotation_conventions.py

"""
Test All Rotation Conventions for DROID cartesian_position

Tests multiple Euler angle conventions (both extrinsic and intrinsic)
to find the exact format used by DROID.
"""

import sys
import numpy as np
import tensorflow_datasets as tfds
from pathlib import Path
import argparse
import pybullet as p
import pybullet_data

# Add parent directory to path
sys.path.append(str(Path(__file__).parent.parent))


def rotation_matrix_to_euler_xyz_intrinsic(R):
    """Extract XYZ Euler angles (intrinsic) from rotation matrix."""
    sy = np.sqrt(R[0, 0]**2 + R[1, 0]**2)
    singular = sy < 1e-6

    if not singular:
        x = np.arctan2(R[2, 1], R[2, 2])
        y = np.arctan2(-R[2, 0], sy)
        z = np.arctan2(R[1, 0], R[0, 0])
    else:
        x = np.arctan2(-R[1, 2], R[1, 1])
        y = np.arctan2(-R[2, 0], sy)
        z = 0

    return np.array([x, y, z])


def euler_to_rotation_matrix(euler, convention='xyz_extrinsic'):
    """
    Convert Euler angles to rotation matrix with various conventions.

    Args:
        euler: [angle1, angle2, angle3] in radians
        convention: Rotation convention
            - 'xyz_extrinsic': X then Y then Z (fixed frame) = Rz*Ry*Rx
            - 'zyx_extrinsic': Z then Y then X (fixed frame) = Rx*Ry*Rz
            - 'xyz_intrinsic': X then Y then Z (moving frame) = Rx*Ry*Rz
            - 'zyx_intrinsic': Z then Y then X (moving frame) = Rz*Ry*Rx

    Returns:
        3x3 rotation matrix
    """
    r1, r2, r3 = euler

    # Individual rotation matrices
    Rx = np.array([
        [1, 0, 0],
        [0, np.cos(r1), -np.sin(r1)],
        [0, np.sin(r1), np.cos(r1)]
    ])

    Ry = np.array([
        [np.cos(r2), 0, np.sin(r2)],
        [0, 1, 0],
        [-np.sin(r2), 0, np.cos(r2)]
    ])

    Rz = np.array([
        [np.cos(r3), -np.sin(r3), 0],
        [np.sin(r3), np.cos(r3), 0],
        [0, 0, 1]
    ])

    if convention == 'xyz_extrinsic':
        # Fixed frame: rotate X, then Y, then Z
        R = Rz @ Ry @ Rx
    elif convention == 'zyx_extrinsic':
        # Fixed frame: rotate Z, then Y, then X
        R = Rx @ Ry @ Rz
    elif convention == 'xyz_intrinsic':
        # Moving frame: rotate X, then Y, then Z
        R = Rx @ Ry @ Rz
    elif convention == 'zyx_intrinsic':
        # Moving frame: rotate Z, then Y, then X
        R = Rz @ Ry @ Rx
    else:
        raise ValueError(f"Unknown convention: {convention}")

    return R


def axis_angle_to_rotation_matrix(axis_angle):
    """Convert axis-angle to rotation matrix using Rodrigues' formula."""
    theta = np.linalg.norm(axis_angle)

    if theta < 1e-6:
        return np.eye(3)

    axis = axis_angle / theta
    K = np.array([
        [0, -axis[2], axis[1]],
        [axis[2], 0, -axis[0]],
        [-axis[1], axis[0], 0]
    ])

    R = np.eye(3) + np.sin(theta) * K + (1 - np.cos(theta)) * (K @ K)
    return R


def quaternion_to_rotation_matrix(quat):
    """Convert quaternion [x, y, z, w] to rotation matrix."""
    x, y, z, w = quat

    R = np.array([
        [1 - 2*(y**2 + z**2), 2*(x*y - w*z), 2*(x*z + w*y)],
        [2*(x*y + w*z), 1 - 2*(x**2 + z**2), 2*(y*z - w*x)],
        [2*(x*z - w*y), 2*(y*z + w*x), 1 - 2*(x**2 + y**2)]
    ])

    return R


def compute_fk_pose(joint_positions):
    """Compute FK to get end-effector pose."""
    client = p.connect(p.DIRECT)
    p.setAdditionalSearchPath(pybullet_data.getDataPath())

    robot_id = p.loadURDF("franka_panda/panda.urdf", useFixedBase=True)

    for i in range(min(7, len(joint_positions))):
        p.resetJointState(robot_id, i, joint_positions[i])

    ee_link_id = 8
    link_state = p.getLinkState(robot_id, ee_link_id)

    position = np.array(link_state[4])
    orientation_quat = np.array(link_state[5])  # [x, y, z, w]

    p.disconnect(client)

    return position, orientation_quat


def test_all_conventions(droid_path, episode_index=0, num_samples=5):
    """Test all rotation conventions."""
    print(f"Loading episode {episode_index}...")

    builder = tfds.builder_from_directory(droid_path)
    dataset = builder.as_dataset(split='train')

    for idx, episode in enumerate(dataset):
        if idx != episode_index:
            continue

        steps = list(episode['steps'])
        print(f"  Total steps: {len(steps)}")

        sample_indices = np.linspace(0, len(steps)-1, num_samples, dtype=int)

        # Test conventions
        conventions = [
            'xyz_extrinsic',
            'zyx_extrinsic',
            'xyz_intrinsic',
            'zyx_intrinsic',
            'axis_angle'
        ]

        results = {conv: [] for conv in conventions}

        print(f"\nTesting {num_samples} frames across {len(conventions)} conventions...")
        print("=" * 80)

        for i, step_idx in enumerate(sample_indices):
            step = steps[step_idx]

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

            # Compute FK
            fk_pos, fk_quat = compute_fk_pose(joint_pos)
            fk_rot = quaternion_to_rotation_matrix(fk_quat)

            # Test position
            pos_error = np.linalg.norm(fk_pos - cart_pos[:3])

            print(f"\nFrame {step_idx}:")
            print(f"  Position error: {pos_error:.6f} m")
            print(f"  Rotation params: {cart_pos[3:]}")
            print(f"  Rotation magnitude: {np.linalg.norm(cart_pos[3:]):.4f}")

            # Test each convention
            for conv in conventions:
                if conv == 'axis_angle':
                    rot_test = axis_angle_to_rotation_matrix(cart_pos[3:])
                else:
                    rot_test = euler_to_rotation_matrix(cart_pos[3:], convention=conv)

                error = np.linalg.norm(rot_test - fk_rot, 'fro')
                results[conv].append(error)
                print(f"  {conv:20s}: {error:.6f}")

        # Summary
        print("\n" + "=" * 80)
        print("SUMMARY - Average Errors")
        print("=" * 80)

        for conv in conventions:
            avg_error = np.mean(results[conv])
            std_error = np.std(results[conv])
            print(f"  {conv:20s}: {avg_error:.6f} (±{std_error:.6f})")

        # Find best
        best_conv = min(conventions, key=lambda c: np.mean(results[c]))
        best_error = np.mean(results[best_conv])

        print(f"\n{'='*80}")
        if best_error < 0.1:
            print(f"✓ BEST MATCH: {best_conv}")
            print(f"  Average error: {best_error:.6f}")
        else:
            print(f"⚠ WARNING: Best match is {best_conv} with error {best_error:.6f}")
            print(f"  This may indicate a coordinate frame transformation")
        print("=" * 80)

        return best_conv, results

    raise ValueError(f"Episode {episode_index} not found")


def main():
    parser = argparse.ArgumentParser(description="Test all rotation conventions for DROID")

    parser.add_argument('--droid-path', type=str,
                       default='/mnt/kevin/data/droid/droid/1.0.0',
                       help='Path to DROID RLDS dataset')
    parser.add_argument('--episode-index', type=int, default=0,
                       help='Episode index to test')
    parser.add_argument('--num-samples', type=int, default=5,
                       help='Number of frames to test')

    args = parser.parse_args()

    print("=" * 80)
    print("DROID Rotation Convention Test")
    print("=" * 80 + "\n")

    try:
        test_all_conventions(args.droid_path, args.episode_index, args.num_samples)
        return 0
    except Exception as e:
        print(f"\n✗ Test failed: {e}")
        import traceback
        traceback.print_exc()
        return 1


if __name__ == "__main__":
    sys.exit(main())