ahanjaya/crossed_arm_point_clouds

Crossed Arm Point Clouds Dataset

This dataset contains 3D point cloud data captured from a LiDAR scanner for crossed arm classification in the context of robot magic trick performance.

Overview

This dataset was collected for training and evaluating the Crossed Arm Voxel Network (CAVN) architecture, a deep learning model designed for 3D point cloud classification in human-robot interaction magic performances. The data supports classification of human arm positions during the interactive magic trick "Which hand is the coin in?", performed by the first fully autonomous humanoid robot magic performance that requires no human assistance during execution.

Dataset Description

The dataset consists of 3D point cloud data captured from participants during crossed-arm poses, where participants hide a coin in one of their hands. The robot must determine which hand contains the coin by analyzing the point cloud data.

Data Structure

The dataset is organized into two main categories:

• left_arm_top/: Point clouds where the left arm is positioned on top during the crossed-arm pose
• right_arm_top/: Point clouds where the right arm is positioned on top during the crossed-arm pose

Each directory contains .npz files (NumPy compressed arrays) with individual point cloud samples.

File Format

Each .npz file contains 3D point cloud data captured from a LiDAR scanner. The files follow the naming convention:

{category}-{sample_number}.npz

For example: left_arm_top-1.npz, right_arm_top-42.npz

Point Cloud Structure:

• Key name: pcl
• Shape: (N, 4) where N varies by sample (typically ~17,000+ points)
• Channels: 4 values per point:
  • Channel 0: X coordinate
  • Channel 1: Y coordinate
  • Channel 2: Z coordinate
  • Channel 3: Intensity/reflectivity value

Usage

To load the point cloud data:

import numpy as np

# Load a single sample
data = np.load('left_arm_top/left_arm_top-1.npz')

# Access the point cloud data
point_cloud = data['pcl']  # Shape: (N, 4)

# The point cloud has 4 channels: x, y, z, intensity
print(f"Point cloud shape: {point_cloud.shape}")  # e.g., (17727, 4)
print(f"Available keys: {data.files}")  # ['pcl']

# Extract individual components
x = point_cloud[:, 0]
y = point_cloud[:, 1]
z = point_cloud[:, 2]
intensity = point_cloud[:, 3]

Citation

If you use this dataset in your research, please cite:

@inproceedings{crossed_arm_magic_trick,
  title={A Crossed Arm Voxel Network for Humanoid Robot-Human Interaction in Magic Performances},
  author={Mandala, Hanjaya and Saeedvand, Saeed and Baltes, Jacky},
  booktitle={[Conference/Journal Name]},
  year={2026},
}

License

This dataset is released under the MIT License.