Simone Arreghini, Nicholas Carlotti, Mirko Nava, Antonio Paolillo, and Alessandro Giusti
Dalle Molle Institute for Artificial Intelligence, USI-SUPSI, Lugano (Switzerland)
Reliable localization of people is fundamental for service and social robots that must operate in close interaction with humans. State-of-the-art human detectors often rely on RGB-D cameras or costly 3D LiDARs. However, most commercial robots are equipped with cameras with a narrow field of view, leaving them unaware of users approaching from other directions, or inexpensive 1D LiDARs whose readings are hard to interpret. To address these limitations, we propose a self-supervised approach to detect humans and estimate their 2D pose from 1D LiDAR data, using detections from an RGB-D camera as supervision. Trained on 70 minutes of autonomously collected data, our model detects humans omnidirectionally in unseen environments with 71% precision, 80% recall, and mean absolute errors of 13 cm in distance and 44◦ in orientation, measured against ground truth data. Beyond raw detection accuracy, this capability is relevant for robots operating in shared public spaces, where omnidirectional awareness of nearby people is crucial for safe navigation, appropriate approach behavior, and timely human-robot interaction initiation using low-cost, privacy-preserving sensing. Deployment in two additional public environments further suggests that the approach can serve as a practical wide-FOV awareness layer for socially aware service robotics.
The paper has been accepted at ARSO 2026. The pdf is available on arXiv.
Cite this work:
@inproceedings{arreghini2026sixth,
title={Sixth-Sense: Self-Supervised Learning of Spatial Awareness of Humans from a Planar Lidar},
author={Arreghini, Simone and Carlotti, Nicholas and Nava, Mirko and Paolillo, Antonio and Giusti, Alessandro},
booktitle={IEEE International Conference on Advanced Robotics and its Social Impacts},
pages = {--},
year={2026}
}
Available on Zenodo - Sixth Sense: Indoor Human Spatial Awareness Dataset.
To install this package, as well as any dependency:
pip install -e .This will automatically install any dependency in your Python environment. Please use a version of Python>=3.10
Download and extract our dataset from Zenodo at this URL.
Copy or make a symlink to the extracted content to the folder /hdf5 such that it directly contains the break_area, corridor, and lab folders.
The file "lidar_human_pose_estimation/config/train_config.yaml" can be used to change the model architecture as well as the loss function used.
A new model can be trained using the command:
python -m lidar_human_pose_estimation.core.train -i <PATH_TO_THE_TRAIN_CONFIG_FILE>
The model will be saved in the model folder of this repository.
The performance of a model on a specific dataset can be tested with:
python -m lidar_human_pose_estimation.core.test -m <PATH_TO_THE_MODEL_FOLDER> -v <PATH_TO_THE_TEST_DATASET> --device cpu
This function will output some model performance metrics as well as create plots.
Depending on what you need to visualize three different Python scripts can be used.
To visualize only the content of an h5 file:
python -m lidar_human_pose_estimation.visualization.vis_h5 -i <PATH_TO_THE_DATASET> -o <PATH_TO_THE_VIDEO_OUTPUT_FOLDER>
To visualize also the optitrack data in a dataset (only for "Lab" environment):
python -m lidar_human_pose_estimation.visualization.vis_h5_optitrack -i <PATH_TO_THE_DATASET> -o <PATH_TO_THE_VIDEO_OUTPUT_FOLDER>
To visualize how a model performs :
python -m lidar_human_pose_estimation.visualization.vis_model -i <PATH_TO_THE_DATASET> -m <PATH_TO_THE_MODEL_FOLDER> --device cpu -o <PATH_TO_THE_VIDEO_OUTPUT_FOLDER>