A low-cost teleoperation system using wearable IMUs to securely and responsively control a robotic arm. Task identification and biometric user authentication are learned from time-series sensor data collected during handling tasks. Across 16 operators and three handling tasks, random forest reaches a macro F1-score of 75.60% for task identification, while logistic regression attains an average Equal Error Rate of approximately 8.89% for user authentication.

This work addresses the challenges of dataset heterogeneity and model adaptability in sensor-based Human Activity Recognition (HAR). We propose a framework that integrates data from multiple inertial sensor datasets through a homogenization pipeline and leverages continual learning strategies to enable models to adapt to new activities without forgetting previously learned ones.

@article{amrani2025leveraging,
  title={Leveraging Dataset Integration and Continual Learning for Human Activity Recognition},
  author={Amrani, Hamza and Micucci, Daniela and Mobilio, Marco and Napoletano, Paolo},
  journal={International Journal of Machine Learning and Cybernetics},
  year={2025},
  doi={10.1007/s13042-024-02456-9},
  publisher={Springer}
}

An end-to-end deep learning framework for decoding open-vocabulary EEG signals into text. The system combines a subject-dependent representation module for raw EEG encoding, a BART language model, and a GPT-4-based evaluation module. Tested on two public ZuCo datasets with 30 subjects, the method achieves a BLEU-1 score of 42.75%, a ROUGE-1-F of 33.28%, and a BERTScore-F of 53.86%, exceeding prior methods by 3.38%, 8.43%, and 6.31%.

@article{amrani2024deep,
  title={Deep Representation Learning for Open Vocabulary Electroencephalography-to-Text Decoding},
  author={Amrani, Hamza and Micucci, Daniela and Napoletano, Paolo},
  journal={IEEE Journal of Biomedical and Health Informatics},
  year={2024},
  doi={10.1109/JBHI.2024.3478609},
  publisher={IEEE}
}

This paper presents a responsive teleoperation system for controlling a robotic arm using wearable inertial sensors. The system maps real-time human arm movements captured by IMU sensors to robotic arm joint commands, enabling intuitive and low-latency remote control for manipulation tasks.

@inproceedings{fornaro2024responsive,
  title={Responsive Teleoperation of a Robotic Arm via Wearable Inertial Sensors},
  author={Fornaro, Alessandro and D'Auria, Daniela and Amrani, Hamza and Napoletano, Paolo},
  booktitle={IEEE CTSoc GEM Conference},
  year={2024},
  organization={IEEE}
}

This paper presents a complete pipeline for EEG-based motor imagery classification targeting robotic control applications. We describe the acquisition protocol using consumer-grade EEG devices, signal processing steps, and deep learning classifiers for decoding motor imagery commands into robotic control signals.

@inproceedings{amrani2024eeg,
  title={EEG Acquisition and Motor Imagery Classification for Robotic Control},
  author={Amrani, Hamza and Micucci, Daniela and Nalin, Marco and Napoletano, Paolo and Rizzi, Ilario},
  booktitle={46th Annual International Conference of the IEEE Engineering in Medicine \& Biology Society (EMBC)},
  year={2024},
  doi={10.1109/EMBC53108.2024.10782895},
  organization={IEEE}
}

We propose an unsupervised deep learning approach for clustering inertial sensor data in Human Activity Recognition. The method combines deep autoencoders with clustering objectives to learn discriminative representations without requiring labeled data, enabling automatic discovery of activity patterns from raw sensor signals.

@inproceedings{amrani2022unsupervised,
  title={Unsupervised Deep Learning-based Clustering for Human Activity Recognition},
  author={Amrani, Hamza and Micucci, Daniela and Napoletano, Paolo},
  booktitle={2022 IEEE 12th International Conference on Consumer Electronics (ICCE-Berlin)},
  pages={1--6},
  year={2022},
  doi={10.1109/ICCE-Berlin56473.2022.9937065},
  organization={IEEE}
}

This paper explores personalized deep learning models for sensor-based Human Activity Recognition. We investigate how user-specific fine-tuning of deep neural networks can significantly improve recognition accuracy compared to generic models, proposing strategies for efficient adaptation with limited per-user labeled data.

@inproceedings{amrani2021personalized,
  title={Personalized Models in Human Activity Recognition using Deep Learning},
  author={Amrani, Hamza and Micucci, Daniela and Napoletano, Paolo},
  booktitle={2020 25th International Conference on Pattern Recognition (ICPR)},
  pages={9682--9688},
  year={2021},
  doi={10.1109/ICPR48806.2021.9412140},
  organization={IEEE}
}