Skip to main content

​
Overview

The UXO Dataset uses a hierarchical transform system to define spatial relationships between coordinate frames. The transform tree consists of two main groups:
  • Setup transforms: Define the sensor mounting configuration (AR3 robot arm, ARIS sonar, GoPro camera)
  • Target transforms: Define the positions of UXO objects in the world frame
All transforms are stored in /calibration/transforms.yaml and can be loaded programmatically using the transform manager.

​
Transform Representation

Each transform consists of:
  • Translation: 3D position offset (x, y, z) in meters
  • Rotation: Quaternion (x, y, z, w) representing orientation
Transforms are defined relative to a parent frame, creating a hierarchical tree structure.

​
Setup Transform Hierarchy

The sensor setup follows this frame hierarchy: 
setup/portal_crane
└── setup/ar3
    ├── setup/aris
    │   └── setup/sonar
    ├── setup/gopro_mounting_hole
    │   └── setup/gopro_mounting_axis
    │       └── setup/camera

​
Portal Crane to AR3 Robot

​
setup/ar3
transform
Defines the AR3 robot arm base relative to the portal crane frame.
frame_id: "setup/ar3"
parent_frame_id: "setup/portal_crane"
translation:
  x: 0.0
  y: 0.174
  z: -0.338
rotation:
  x: 0.0
  y: 1.0
  z: 0.0
  w: 0.0
The AR3 is offset 17.4 cm in Y and -33.8 cm in Z from the portal crane, with a 180° rotation around the Y-axis.

​
AR3 to ARIS Sonar

frame_id: "setup/aris"
parent_frame_id: "setup/ar3"
translation:
  x: 0.0
  y: 0.0
  z: 0.0
rotation:
  x: 0.0
  y: 0.0
  z: 0.0
  w: 1.0
The ARIS base frame is coincident with the AR3 end-effector.

​
AR3 to GoPro Camera

The GoPro camera mount has an intermediate frame hierarchy: 
frame_id: "setup/gopro_mounting_hole"
parent_frame_id: "setup/ar3"
translation:
  x: 0.1265
  y: 0.023
  z: -0.1055
rotation:
  x: 0.0
  y: 0.0
  z: 0.0
  w: 1.0
Reference point for the GoPro mounting bracket.
frame_id: "setup/gopro_mounting_axis"
parent_frame_id: "setup/gopro_mounting_hole"
translation:
  x: 0.02125
  y: 0.0
  z: -0.0075
rotation:
  x: 0.0
  y: 0.0
  z: 0.0
  w: 1.0
Intermediate frame for mounting axis adjustments.
frame_id: "setup/camera"
parent_frame_id: "setup/gopro_mounting_axis"
translation:
  x: 0.01925
  y: -0.0195
  z: -0.0415
rotation:
  x: 0.5
  y: 0.5
  z: 0.5
  w: 0.5
Final camera optical center frame with standard camera orientation.

​
Target Transforms

All target objects are positioned relative to the world frame. The dataset includes five UXO targets: 
frame_id: "target/100lbs"
parent_frame_id: "world"
translation:
  x: 1.251079
  y: 1.592704
  z: -1.489828
rotation:
  x: 0.0
  y: 0.0
  z: 0.0
  w: 1.0
Position: (1.25m, 1.59m, -1.49m)

​
Using the Transform Manager

The dataset provides a Python utility to load and work with transforms using the pytransform3d library.

​
Loading Transforms

import numpy as np
from pytransform3d.transform_manager import TransformManager

# Import from the exported dataset structure
from dataset.calibration.tf_demo.transforms import get_tf_manager

# Or if working directly with the source repository
# from demo.transforms import get_tf_manager

# Load the complete transform tree
tm = get_tf_manager()

​
Computing Transforms

The transform manager automatically computes transformations between any two frames: 
# Get transform from camera to a target
camera_to_target = tm.get_transform("setup/camera", "target/100lbs")

# Get transform from sonar to a target
sonar_to_target = tm.get_transform("setup/sonar", "target/mortar_shell")

# Get transform between sensors
camera_to_sonar = tm.get_transform("setup/camera", "setup/sonar")

​
Transform Format

Transforms are represented as 4x4 homogeneous transformation matrices: 
# Transform matrix structure
# [[R11, R12, R13, tx],
#  [R21, R22, R23, ty],
#  [R31, R32, R33, tz],
#  [  0,   0,   0,  1]]
Where the 3x3 rotation matrix is in the upper-left, and the translation vector is in the right column.

​
Internal Implementation

The transform loading process (from /workspace/source/demo/transforms.py:22-28): 
for tf in tf_statics[object]:
    transform = pt.transform_from_pq(
                    np.hstack((
                        np.array([tf['translation']['x'], tf['translation']['y'], tf['translation']['z']]), 
                        np.array([tf['rotation']['w'], tf['rotation']['x'], tf['rotation']['y'], tf['rotation']['z']]))))
    
    tm.add_transform(tf['frame_id'], tf['parent_frame_id'], transform)
The quaternion order in the YAML file is (x, y, z, w), but pytransform3d expects (w, x, y, z) format.

​
Coordinate Frame Conventions

  • World Frame: Fixed reference frame, typically aligned with the pool structure
  • Setup Frames: Moving frames attached to the robotic system
  • Camera Frame: Follows standard computer vision convention (Z-forward, X-right, Y-down)
  • Sonar Frame: Aligned with the ARIS imaging plane

​
File Location

The transforms are stored in: 
/calibration/transforms.yaml
The Python loader utility is located at: 
/demo/transforms.py