Back to Home

All Publications

📅 2026 ICRA 2026

A Modular, Wireless and Wearable Biosignal Acquisition Platform

Authors: Antonios Doukakis, Aikaterini Smyrli, Makis Livadas, Henrique De Melo Ribeiro, Shadiya Alingal-Meethal, Mohamad Reza Shahabian Alashti, Gabriella Lakatos, Patrick Holthaus, Farshid Amirabdollahian
Published In ICRA 2026
Year 2026
Abstract
We present a modular, wireless biosignal acquisition platform designed to enable scalable electromyography (EMG) and inertial measurement unit (IMU) sensing for wearable robotics applications. The system supports up to 64 EMG channels and integrates a 9-axis IMU, leveraging a distributed Leader-Follower board architecture. In this work, we demonstrate synchronised acquisition of 32 EMG channels together with IMU motion data in a fully wireless setup. The embedded firmware ensures low-latency, high-fidelity streaming at 1.4 kHz over a 2.4-GHz industrial, scientific and medical (ISM) band link. Benchmarking shows that the platform maintains uniformly strong performance across noise, power, footprint, bandwidth, and scalability, in contrast to existing designs that optimize only a single metric. Experimental demonstrations confirm reliable acquisition of high-density EMG and IMU signals across functional activities, highlighting the device’s robustness and wearability. The proposed system provides a compact and flexible solution for intent-aware wearable technologies, with applications in assistive exosuits, rehabilitation, and human–robot interaction.
View Paper

Official page: University of Hertfordshire Research Profiles

Accepted for publication at the 2026 IEEE International Conference on Robotics and Automation (ICRA), Vienna, Austria, in press.

This work presents a comprehensive platform for wireless biosignal acquisition, designed to support research in wearable robotics and human-machine interaction.

📅 2026 SSRN Preprint

A Deployment-Oriented Validation Framework for Cross-Day Wearable Activity Recognition Using EMG and IMU

Authors: Mohamad Reza Shahabian Alashti, Shadiya Alingal Meethal, Patrick Holthaus, Gabriella Lakatos, Farshid Amirabdollahian
Published In SSRN Preprint
Year 2026
Abstract
Background and Objective: Evaluation protocols in biomedical machine learning can substantially overestimate real-world performance when training and testing conditions do not reflect deployment variability. This issue is especially important for multi-day wearable sensing, where sensor reattachment and day-wise shifts can alter signal characteristics and mislead model selection. This study proposes a deployment-oriented validation framework for biomedical wearable computing. It applies it to lower-limb activity recognition using high-density EMG (HD-EMG), bipolar EMG (BP-EMG), and inertial measurement unit (IMU) sensing. Methods: Using the public multi-day MyPredict3 dataset, we combined pooled offline testing with strict leave-one-day-out separation and chronological sequential inference on both an included day and a fully held-out day. Modalities were evaluated under matched preprocessing, model capacity, and training conditions. Sensor configuration, HD-EMG coverage, and overlapping-window aggregation were examined as secondary analyses. Results: Conventional pooled testing suggested strong performance across modalities (92.5% bilateral BP-EMG, 96.9% bilateral IMU, and 85.2% 64-channel HD-EMG). In contrast, deployment-oriented sequential testing revealed substantially different robustness profiles. Bilateral IMU showed the smallest cross-day loss (1.7 percentage points), BP-EMG showed moderate degradation (7.8–11.5 percentage points, improved by bilateral sensing), and HD-EMG showed the largest degradation (23.3–23.7 percentage points) with little benefit from increased grid density. Overlap and temporal aggregation improved same-day stability but did not materially reduce the cross-day performance gap. Conclusions: The proposed framework makes deployment-relevant failure modes measurable and provides a reusable validation protocol for biomedical wearable computing. Within this controlled setting, sensing modality was the dominant determinant of longitudinal robustness, indicating that reliable EMG-based deployment will likely require calibration-efficient or adaptive strategies.
View Paper

Preprint: SSRN

Submitted to Computer Methods and Programs in Biomedicine (CMPB). This preprint has not yet been peer reviewed.

This work shows that conventional pooled train/test splits can substantially overstate real-world performance for multi-day wearable activity recognition. Using strict leave-one-day-out and chronological sequential evaluation on the MyPredict3 dataset, IMU sensing proved far more robust to cross-day distribution shift than EMG, motivating calibration-efficient or adaptive strategies for reliable EMG-based deployment in assistive wearable robotics.

📅 2026 ICSR+Art 2026

EMG-Based Lower Limb Activity Recognition for Exoskeleton-Assisted and Unassisted Locomotion Using Deep Learning

Authors: Shadiya Alingal-Meethal, Mohamad Reza Shahabian Alashti, Martina Mosso, Massimo Sartori, Mohamed Irfan Refai, Ilaria Pacifico, Patrick Holthaus, Gabriella Lakatos, Farshid Amirabdollahian
Published In ICSR+Art 2026
Year 2026
Abstract
With advancements in exoskeleton technology, the integration of advanced machine learning techniques with sensor data can essentially improve the accuracy and effectiveness of activity detection. This paper explores the potential of machine learning algorithms to detect activities with the help of features extracted from Electromyogram (EMG) signals at a window length of as small as 100 ms, with and without an exoskeleton. Accurate activity recognition is also important for social and assistive robots, enabling more adaptive human–robot interaction and context-aware assistance in rehabilitation, healthcare, and occupational assistance environments. Electromyogram signals were collected from trunk, abdominal and thigh muscles of healthy participants. A one-dimensional convolutional neural network (CNN) was employed for detecting three classes: no activity, walking and walking with weight. Our evaluation shows that the CNN model was able to detect the activities with similar accuracy in the presence and absence of an exoskeleton. An average accuracy of 85.66% was observed for the condition without the exoskeleton, and an average accuracy of 86% was observed for the exoskeleton-assisted condition. Also, when the CNN model was trained on exoskeleton-assisted data and tested on EMG data without exoskeleton (and vice versa), a notable reduction in accuracy was observed. This suggests that the machine learning model performs reliably under in-domain evaluation and struggles to generalise across conditions. This highlights the need for domain adaptation to mitigate the exoskeleton-induced distribution shift in EMG.
View Paper

Official page: University of Hertfordshire Research Profiles

Accepted for publication at the 18th International Conference on Social Robotics + Art (ICSR+Art 2026), London, UK.

This work benchmarks a 1D CNN for recognising walking activities from trunk, abdominal, and thigh EMG signals, evaluating how well the model generalises between exoskeleton-assisted and unassisted locomotion and motivating domain adaptation for exoskeleton-induced distribution shift in EMG-based intent detection.

📅 2026 IEEE Trans. Multimedia (submitted)

HAR-Agent: Multilingual Multimodal Activity Recognition via Knowledge-Distilled LLM Reasoning

Authors: Khashayar Ghamati, Mohammad Reza Shahabian Alashti, Ali Fallahirahmatabadi, Abolfazl Zaraki
Published In IEEE Trans. Multimedia (submitted)
Year 2026
Abstract
Building intelligent agents capable of understanding human activities requires reasoning modules that can interpret complex visual and auditory observations. We present HAR-Agent, a multilingual multimodal human activity recognition system whose core contribution is a methodology for developing domain-adapted LLM reasoning through fine-tuning on real-world activity data and compressing it via knowledge distillation for deployment. Using the RHM-HAR dataset (6,701 robot-mounted video clips; 14 activity classes) for adaptation, we instruction-tune a 72B-parameter teacher (Qwen2.5-72B-Instruct) and compare instruction tuning (IT) to supervised fine-tuning (SFT) across 17 model configurations (1.5B–72B). IT outperforms SFT at every scale (17–28 point advantage), and distilled students down to 1.5B parameters enable inference on consumer hardware. We evaluate in-domain on the held-out RHM-HAR validation split and cross-domain on Toyota Smarthome to quantify transfer beyond the training distribution. The architecture unifies vision (LLaVA), audio (Whisper), and direct text pathways into a common textual representation consumed by the reasoning module. On a controlled multilingual audio benchmark spanning five language variants, the audio pathway achieves 89.2% accuracy with sub-second latency. We further introduce deployment-centric metrics to quantify knowledge transfer and hardware accessibility.

Submitted to IEEE Transactions on Multimedia (2026), currently under review.

HAR-Agent turns a large multimodal reasoning model into something deployable: a 72B-parameter teacher is instruction-tuned on real-world activity data and distilled across 17 model configurations, with students as small as 1.5B parameters running on consumer hardware. Vision (LLaVA), audio (Whisper), and text pathways are unified into a common textual representation, and a five-language audio benchmark (Chinese, Spanish, Farsi, US/UK English) validates the multilingual pathway at 89.2% accuracy with sub-second latency. The paper also introduces deployment-centric metrics for knowledge transfer and hardware accessibility.

📅 2026 ICSR+Art 2026

Pexformer: Robust Indoor Human Localisation via Patch-level Tokenisation and Semi-Permeable Attention

Authors: Baobing Zhang, Sehrish Rafique, Mohamad Reza Shahabian Alashti, Shadiya Alingal-Meethal, Vignesh Velmurugan, Patrick Holthaus, Gabriella Lakatos, Angela Dickinson, Farshid Amirabdollahian
Published In ICSR+Art 2026
Year 2026
Abstract
Social robots are increasingly explored within Ambient Assisted Living (AAL) settings, to support people living alone in ways that preserve autonomy and dignity. To facilitate effective interactions and prompt assistance, such as proactive check-ins, safe navigation, or escalation when something appears wrong, knowing the person's location in the home is invaluable. Non-intrusive and exteroceptive sensors are often used to estimate a person's location in a house. However, existing data-driven methods often struggle with extreme class imbalance characterised by long-tail distributions of room occupancy and the inherent noise of sparse sensor triggers emerging from AAL settings. To address these challenges, this paper introduces Patch-Excel-Transformer (Pexformer), a novel architecture that adapts efficient computational paradigms from the tabular domain to time-series localisation tasks, to maintain accuracy with such sparse data. Pexformer leverages patch-level tokenisation to effectively capture local temporal dynamics and integrates a Semi-Permeable Attention (SPA) mechanism to construct hierarchical feature interactions, reducing computational complexity while preserving critical information. Notably, we further observe that a simple random permutation of tokens acts as an effective regulariser and performs comparably to or better than mutual information (MI)-based ordering, avoiding the need for costly statistical pre-computation. Experiments on real-world smart home datasets confirm that Pexformer achieves state-of-the-art localisation accuracy and strong balanced performance, particularly for under-represented room categories, without relying on complex oversampling techniques.
View Paper

Accepted for publication at the 18th International Conference on Social Robotics + Art (ICSR+Art 2026), London, UK. Lecture Notes in Computer Science, Springer, in press.

Preprint: PDF

Code: GitHub - pexformer

Pexformer adapts tabular-domain computational paradigms to time-series indoor localisation, combining patch-level tokenisation with a Semi-Permeable Attention mechanism to handle the extreme class imbalance and sparse sensor triggers typical of Ambient Assisted Living settings. On real-world smart home datasets it substantially outperforms BiLSTM, NODE, TabTransformer, and DCN V2 baselines (93.3% accuracy, 83.52% Macro-F1), with a simple random-permutation regularisation strategy proving as effective as costlier mutual-information-based token ordering.

📅 2026 ICSR+Art 2026

From Pilot Data to Protocol: Sample-Size Guidance for Multimodal Intent Detection in Assistive Wearable Robotics

Authors: Mohamad Reza Shahabian Alashti, Shadiya Alingal-Meethal, Patrick Holthaus, Gabriella Lakatos, Farshid Amirabdollahian
Published In ICSR+Art 2026
Year 2026
Abstract
Designing a sensible data-collection protocol for intent detection in assistive exoskeletons and wearable robots is challenging: too little data yields unstable models, while collecting 'as much data as possible' is costly and often unnecessary. This challenge is particularly relevant in social robotics and human-robot interaction, where HAR from biosignals such as EMG is used to infer user intent and physical state for safer, more adaptive assistance. We propose a generic, model-agnostic procedure that turns a small pilot recording into concrete sample-size guidance using learning curves. Using a 32-channel high-density EMG (HD-EMG) grid on the thigh and a co-located IMU, we record eight repetitions of six lower-limb activities relevant to locomotion assistance and extract 100 ms windows. We compare RF, SVM, LDA, and a ResNet-18 CNN under leave-one-trial-out (LOTO) evaluation, estimating the training fraction needed to reach 90% of each model's peak accuracy and the plateau where adding 10% more data yields less than 1 percentage-point gain. On this pilot dataset, RF/SVM/LDA typically meet both criteria after 20–30% of available windows, whereas the CNN continues to improve up to approximately 70–90%. IMU features outperform HD-EMG alone, and EMG+IMU fusion achieves the highest accuracy. Overall, the protocol provides per-model sample targets and a principled stopping rule to reduce recording and recalibration burden in data-efficient exoskeleton intent detection for human augmentation and wellbeing.
View Paper

Accepted for publication at the 18th International Conference on Social Robotics + Art (ICSR+Art 2026), London, UK. Lecture Notes in Computer Science, Springer, in press.

Preprint: PDF

This paper proposes a model-agnostic, learning-curve-based procedure for turning a small pilot recording into concrete sample-size targets for multimodal (HD-EMG + IMU) intent detection, giving practitioners a principled stopping rule that reduces recording and recalibration burden when designing data-collection protocols for assistive wearable robotics.

📅 2026 ICSR+Art 2026

Vision–Language Models for Fall Detection in Socially Assistive Robotics: Zero-Shot Prompting and Few-Shot Calibration

Authors: Mohamad Reza Shahabian Alashti, Khashayar Ghamati, Abolfazl Zaraki, Baobing Zhang, Patrick Holthaus, Shadiya Alingal-Meethal, Vignesh Velmurugan, Gabriella Lakatos, Angela Dickinson, Farshid Amirabdollahian
Published In ICSR+Art 2026
Year 2026
Abstract
Vision–language models (VLMs) offer a promising route to fall detection for socially assistive robots in home and care settings, where timely recognition can trigger assistance or further verification by carers or interactive robots. Most vision-based fall detectors are supervised and require task-specific labelled data and/or robust pose estimation, which can be brittle under occlusion and viewpoint changes and costly to adapt across deployments. This paper investigates whether pretrained VLMs can enable data-free fall detection via zero-shot prompting, and how much a lightweight few-shot calibration step improves performance without requiring backbone tuning. We present (i) a zero-shot detector based on a balanced contrastive prompt bank, and (ii) a few-shot variant that trains only a linear classifier on frozen VLM embeddings. We evaluate these alongside three skeleton-based supervised baselines (2D CNN, 3D CNN, ViT) and a rule-based heuristic on a balanced test set of 40 single-person videos (20 fall, 20 non-fall), with identical windowing (32 frames, 50% overlap) and video-level aggregation (majority vote). The few-shot VLM achieves 100% accuracy, while the zero-shot VLM reaches 92.5% accuracy without fall-specific training data (3 false positives on non-fall videos). Skeleton-based baselines achieve 97.5–100% accuracy but require pose extraction, increasing pipeline complexity. These results suggest that pretrained VLMs can provide a practical perception trigger for robot-in-the-loop verification and escalation in assistive care, with zero-shot prompting achieving high recall at the cost of a small number of false alarms.
View Paper Watch Video

Accepted for publication at the 18th International Conference on Social Robotics + Art (ICSR+Art 2026), London, UK. Lecture Notes in Computer Science, Springer, in press.

Preprint: PDF

Demo video: YouTube

Code: GitHub - FallDetection-public

This work compares zero-shot and few-shot vision–language model prompting against skeleton-based supervised baselines (2D/3D CNN, Vision Transformer, ST-GCN, TCN+Transformer) and a rule-based heuristic for fall detection, served through a Flask web app for interactive demonstration. The few-shot VLM approach reaches 100% accuracy on a balanced 40-video test set without requiring backbone fine-tuning, making it a practical, data-efficient perception trigger for robot-in-the-loop verification in socially assistive care settings.

📅 2025 ICSR 2025

Towards Memory-Driven Agentic AI for Human Activity Recognition

Authors: Mohamad Reza Shahabian Alashti, Khashayar Ghamati, Hooman Samani, Abolfazl Zaraki
Published In ICSR 2025
Year 2025
Abstract
This paper proposes a novel, scalable agentic AI architecture designed to enhance human activity recognition across data modalities by embedding memory-driven reasoning and context awareness. The architecture integrates multimodal sensing, deliberative reasoning through supervised learning and context-aware language models, and memory mechanisms, including short-term memory for tracking immediate activity transitions and long-term memory for embedding experiential knowledge. The evaluation of the proposed model using two major datasets namely RHM (6.7K video clips of 14 known activities) and Toyota Smart Home (16K video clips of 31 unknown activities) demonstrates significant improvements, achieving 60% accuracy when combining contextual information with supervised model output, compared to 40% accuracy with context alone and 35% with supervised models on unseen data. By overcoming the limitations of traditional HAR approaches, this research advances the development of responsive and intelligent robotic systems, facilitating more natural and effective human-robot collaboration.
Watch Video

Official page: University of Hertfordshire Research Profiles

Talk video: YouTube

Accepted for publication at the International Conference on Social Robotics (ICSR) 2025.

This paper introduces novel concepts from agentic AI to human activity recognition, proposing memory-driven systems that can adapt their behavior based on context and user history.

📅 2024 BioRob 2024

Efficient Skeleton-based Human Activity Recognition in Ambient Assisted Living Scenarios with Multi-view CNN

Authors: Mohamad Reza Shahabian Alashti, Mohammad Bamorovat Abadi, Patrick Holthaus, Catherine Menon, Farshid Amirabdollahian
Published In BioRob 2024
Year 2024
Abstract
Human activity recognition (HAR) plays a critical role in diverse applications and domains, from assessments of ambient assistive living (AAL) settings and the development of smart environments to human-robot interaction (HRI) scenarios. However, using mobile robot cameras in such contexts has limitations like restricted field of view and possible noise. Therefore, employing additional fixed cameras can enhance the field of view and reduce susceptibility to noise. Nevertheless, integrating additional camera perspectives increases complexity, a concern exacerbated by the number of real-time processes that robots should perform in the AAL scenario. This paper introduces our methodology that facilitates the combination of multiple views and compares different aspects of fusing information at low, medium and high levels. Their comparison is guided by parameters such as the number of training parameters, floating-point operations per second (FLOPs), training time, and accuracy. Our findings uncover a paradigm shift, challenging conventional beliefs by demonstrating that simplistic CNN models outperform their more complex counterparts using this innovation. Additionally, the pivotal role of pipeline and data combination emerges as a crucial factor in achieving better accuracy levels. In this study, integrating the additional view with the Robot-view resulted in an accuracy increase of up to 25 %. Ultimately, we have successfully attained a streamlined and efficient multi-view HAR pipeline, which will now be incorporated into AAL interaction scenarios.
View Paper Watch Video

Official page: IEEE Xplore

Talk video: YouTube

Published at the 10th IEEE RAS/EMBS International Conference for Biomedical Robotics and Biomechatronics (BioRob).

This work demonstrates how multi-view skeleton data can be effectively processed using convolutional neural networks for real-time activity recognition in smart home environments.

📅 2024 BioRob 2024

Robotic Vision and Multi-View Synergy: Action and Activity Recognition in Assisted Living Scenarios

Authors: Mohammad Bamorovat Abadi, Mohamad Reza Shahabian Alashti, Patrick Holthaus, Catherine Menon, Farshid Amirabdollahian
Published In BioRob 2024
Year 2024
Abstract
The significance of Human-Robot Interaction (HRI) is increasingly evident when integrating robotics within human-centric settings. A crucial component of effective HRI is Human Activity Recognition (HAR), which is instrumental in enabling robots to respond aptly in human presence, especially within Ambient Assisted Living (AAL) environments. Since robots are generally mobile and their visual perception is often compromised by motion and noise, this paper evaluates methods by merging the robot's mobile perspective with a static viewpoint utilising multi-view deep learning models. We introduce a dual-stream Convolutional 3D (C3D) model to improve vision-based HAR accuracy for robotic applications. Utilising the Robot House Multiview (RHM) dataset, which encompasses a robotic perspective along with three static views (Front, Back, Top), we examine the efficacy of our model and conduct comparisons with the dual-stream ConvNet and Slow-Fast models. The primary objective of this study is to enhance the accuracy of robot viewpoints by integrating them with static views using dual-stream models. The metrics for evaluation include Top-1 and Top-5 accuracy. Our findings reveal that the integration of static views with robotic perspectives significantly boosts HAR accuracy in both Top-1 and Top-5 metrics across all models tested. Moreover, the proposed dual-stream C3D model demonstrates superior performance compared to the other contemporary models in our evaluations.
View Paper

Official page: IEEE Xplore

Published at the 10th IEEE RAS/EMBS International Conference for Biomedical Robotics and Biomechatronics (BioRob).

This paper explores how multiple camera viewpoints can be integrated with robotic perception systems to achieve more reliable activity recognition for elderly care applications.

📅 2023 ACHI 2023

Lightweight Human Activity Recognition for Ambient Assisted Living

Authors: Mohamad Reza Shahabian Alashti, Mohammad Bamorovat Abadi, Patrick Holthaus, Catherine Menon, Farshid Amirabdollahian
Published In ACHI 2023
Year 2023
Abstract
Ambient assisted living (AAL) systems aim to improve the safety, comfort, and quality of life for the populations with specific attention given to prolonging personal independence during later stages of life. Human activity recognition (HAR) plays a crucial role in enabling AAL systems to recognise and understand human actions. Multi-view human activity recognition (MV-HAR) techniques are particularly useful for AAL systems as they can use information from multiple sensors to capture different perspectives of human activities and can help to improve the robustness and accuracy of activity recognition. In this work, we propose a lightweight activity recognition pipeline that utilizes skeleton data from multiple perspectives to combine the advantages of both approaches and thereby enhance an assistive robot's perception of human activity. The pipeline includes data sampling, input data type, and representation and classification methods. Our method modifies a classic LeNet classification model (M-LeNet) and uses a Vision Transformer (ViT) for the classification task. Experimental evaluation on a multi-perspective dataset of human activities in the home (RH-HAR-SK) compares the performance of these two models and indicates that combining camera views can improve recognition accuracy. Furthermore, our pipeline provides a more efficient and scalable solution in the AAL context, where bandwidth and computing resources are often limited.
View Paper

Official page: University of Hertfordshire Research Profiles

This work is published at the Sixteenth International Conference on Advances in Computer-Human Interactions and the focuses on model compression and optimization techniques to enable real-time HAR on resource-constrained devices commonly found in smart home environments.

Key contributions include:

  • Efficient network architectures for edge deployment
  • Knowledge distillation for model compression
  • Real-time performance on embedded devices
  • Maintained accuracy with reduced computational requirements
📅 2023 ACHI 2023

RHM: Robot House Multi-view Human Activity Recognition Dataset

Authors: Mohammad Bamorovat Abadi, Mohamad Reza Shahabian Alashti, Patrick Holthaus, Catherine Menon, Farshid Amirabdollahian
Published In ACHI 2023
Year 2023
Abstract
With the recent increased development of deep neural networks and dataset capabilities, the Human Action Recognition (HAR) domain is growing rapidly in terms of both the available datasets and deep models. Despite this, there are some lacks at datasets specifically covering the Robotics field and Human-Robot interaction. We prepare and introduce a new multi-view dataset to address this. The Robot House Multi-View dataset (RHM) contains four views: Front, Back, Ceiling, and Robot Views. There are 14 classes with 6701 video clips for each view, making a total of 26804 video clips for the four views. The lengths of the video clips are between 1 to 5 seconds. The videos with the same number and the same classes are synchronized in different views. In the second part of this paper, we consider how single streams afford activity recognition using established state-of-the-art models. We then assess the affordance for each of the views based on information theoretic modelling and mutual information concept. Furthermore, we benchmark the performance of different views, thus establishing the strengths and weaknesses of each view relevant to their information content and performance of the benchmark. Our results lead us to conclude that multi-view and multi-stream activity recognition has the added potential to improve activity recognition results.
View Paper

Official page: University of Hertfordshire Research Profiles

Dataset page: Robot House Multiview Human Activity Recognition Dataset

A comprehensive RGB video dataset including:

  • Multiple camera viewpoints covering entire living spaces
  • Natural activities performed in realistic home settings
  • Long-duration recordings capturing activity variations
  • Comprehensive annotations and metadata
  • Suitable for various computer vision tasks

The Robot House provides a unique realistic testbed for ambient assisted living research, and this dataset captures genuine human behaviors in that environment.

📅 2023 ACHI 2023

RHM-HAR-SK: A multi-view dataset with skeleton data for Ambient Assisted Living Research

Authors: Mohamad Reza Shahabian Alashti, Mohammad Bamorovat Abadi, Patrick Holthaus, Catherine Menon, Farshid Amirabdollahian
Published In ACHI 2023
Year 2023
Abstract
Ambient assisted living (AAL) systems aim to improve the safety, comfort, and quality of life for the populations with specific attention given to prolonging personal independence during later stages of life. Human activity recognition (HAR) plays a crucial role in enabling AAL systems to recognise and understand human actions. Multi-view human activity recognition (MV-HAR) techniques are particularly useful for AAL systems as they can use information from multiple sensors to capture different perspectives of human activities and can help to improve the robustness and accuracy of activity recognition. In this work, we propose a lightweight activity recognition pipeline that utilizes skeleton data from multiple perspectives to combine the advantages of both approaches and thereby enhance an assistive robot's perception of human activity. The pipeline includes data sampling, input data type, and representation and classification methods. Our method modifies a classic LeNet classification model (M-LeNet) and uses a Vision Transformer (ViT) for the classification task. Experimental evaluation on a multi-perspective dataset of human activities in the home (RH-HAR-SK) compares the performance of these two models and indicates that combining camera views can improve recognition accuracy. Furthermore, our pipeline provides a more efficient and scalable solution in the AAL context, where bandwidth and computing resources are often limited.
View Paper

Official page: University of Hertfordshire Research Profiles

A dataset page is available here: Robot House RHM-HAR-SK

A publicly available dataset featuring:

  • Multi-view synchronized skeleton sequences
  • Diverse activities relevant to elderly care
  • Multiple subjects with varied demographics
  • High-quality 3D pose annotations
  • Benchmark evaluation protocols

The dataset has been used by multiple research groups for developing and evaluating HAR algorithms in assisted living contexts.

📅 2022 Alan Turing Institute Report

Data augmentation and synthetic data generation for low-frequency and sparse data problems

Authors: Mohamad Reza Shahabian Alashti, Alan Turing Institute Team, AMRC Collaborators
Published In Alan Turing Institute Report
Year 2022
Abstract
The Advanced Manufacturing Research Centre (AMRC) group is part of the UK’s High Value Manufacturing (HVM) Catapult, whose mission is to accelerate the concepts-to-commercial-reality process and create a sustainable future for high-value manufacturing. Many manufacturers rely on the manufacture of a small number of high-value workpieces. In contrast to high-volume production, any workpieces rejected in high-value manufacturing represent a large individual investment of resources. Nonetheless, the reasons for rejection are often difficult to determine, which hinders the improvement of the manufacturing process. The reason for this difficulty is the lack of data available on the manufactured workpiece - due to it undergoing several processes, the scarcity of data collected, and the limited sample size in low-volume, high-value manufacturing. Early prediction of workpiece failure would increase productivity and reduce waste by an early stop of the manufacturing process. The workpieces of interest for this Data Study Group are high-value, low-yield ones since they are either made from expensive materials or undergo expensive treatments. Thus, any improvements that lead to fewer rejected pieces will be of high business value and save resources. AMRC has recently collected data on 16 such products, from their forging through machining to their final quality check. This challenge aims to explore the potential of this novel dataset for high-value, low-yield research with a particular emphasis on failure prediction, data augmentation, and data viability.
View Paper

Technical report from the Data Study Group collaboration with the Alan Turing Institute and Advanced Manufacturing Research Centre.

Key contributions:

  • Novel augmentation strategies for sparse datasets
  • Synthetic data generation preserving statistical properties
  • Validation approaches for augmented data
  • Application to real-world manufacturing challenges
  • Guidelines for practitioners working with limited data
📅 2021 4th UKRAS21 Conference: Robotics at home Proceedings

Human activity recognition in RoboCup@ home: Inspiration from online benchmarks

Authors: Mohamad Reza Shahabian Alashti, Mohammad Bamorovat Abadi, Patrick Holthaus, Catherine Menon, Farshid Amirabdollahian
Published In 4th UKRAS21 Conference: Robotics at home Proceedings
Year 2021
Abstract
Human activity recognition is an important aspect of many robotics applications. In this paper, we discuss how well the RoboCup@home competition accounts for the importance of such recognition algorithms. Using public benchmarks as an inspiration, we propose to add a new task that specifically tests the performance of human activity recognition in this league. We suggest that human-robot interaction research in general can benefit from the addition of such a task as RoboCup@home is considered to accelerate, regulate, and consolidate the field.
View Paper

Official page: University of Hertfordshire Research Profiles

This work established standardized benchmarks and evaluation protocols for HAR in home environments, facilitating fair comparison across different approaches and promoting reproducible research.

📅 2021 4th UKRAS21 Conference: Robotics at home Proceedings

Affordable Robot Mapping using Omnidirectional Vision

Authors: Mohammad Bamorovat Abadi, Mohamad Reza Shahabian Alashti, Patrick Holthaus, Catherine Menon, Farshid Amirabdollahian
Published In 4th UKRAS21 Conference: Robotics at home Proceedings
Year 2021
Abstract
Mapping is a fundamental requirement for robot navigation.In this paper, we introduce a novel visual mapping method that relies solely on a single omnidirectional camera.We present a metric that allows us to generate a map from the input image by using a visual Sonar approach.The combination of the visual sonars with the robot's odometry enables us to determine a relation equation and subsequently generate a map that is suitable for robot navigation.Results based on visual map comparison indicate that our approach is comparable with the established solutions based on RGB-D cameras or laser-based
View Paper

Official page: UH Research Archive (UHRA)

This work demonstrates how affordable omnidirectional cameras can be used for simultaneous localization and mapping (SLAM), reducing the cost barrier for robotics research and education.

📅 2018 6th RSI International Conference on Robotics and Mechatronics (IcRoM)

Automatic ROI Detection in Lumbar Spine MRI

Authors: Mohamad Reza Shahabian Alashti, Mohammad Reza Daliri, Behnam Jamei
Published In 6th RSI International Conference on Robotics and Mechatronics (IcRoM)
Year 2018
Abstract
Low back pain (LBP) is one of the most common diseases affecting a large number of people. Diagnosis and treatment of LBP require quick, accurate imaging methods. Magnetic resonance imaging (MRI) is effective in distinguishing between vertebra, intervertebral disc and spinal cord, and thus is used frequently in spinal cord injury (SCI) diagnosis. This paper proposes a fully automated approach to detecting region of interest (ROI) using T2-weighted MRI images. Our dataset included the cases of 100 patients who suffered from LBP. In total, 2000 axial and 1200 sagittal ROI were marked in the Lumbar spine. Extracted ROIs were used in the cascade classifier learner. In this method, ROI detection consists of two processes. First the ROIs are specified using the cascade classifier, and then via a process, non-regions of interest (NROIs) are discarded. Histogram of Oriented Gradient (HOG) was used as the feature descriptor in each stage of the Cascade classifier. This method does not require background knowledge of input images and it is reliable regardless of the images size, contrast and clinical abnormally of cases. The quantitative and qualitative evaluation results of the proposed ROI detector were 83% and above 94%, respectively.
View Paper

Official page: IEEE Xplore

This work from the MSc research applies machine learning and image processing techniques to automatically identify relevant anatomical structures in medical imaging, assisting radiologists in diagnosis and treatment planning.

📅 2017 5th RSI International Conference on Robotics and Mechatronics (ICRoM)

FARAT1: An Upper Body Exoskeleton Robot

Authors: Farzad Cheraghpour, Farbod Farzad, Milad Shahbabai, Mohamad Reza Shahabian Alashti
Published In 5th RSI International Conference on Robotics and Mechatronics (ICRoM)
Year 2017
Abstract
PExoskeleton robots were designed to increase strength and endurance of human limbs. This kind of robots could be used to increase the physical ability of either disabled or ordinary people for executing motion or manipulation tasks. The important point is to design such a shape that could be used safely, and accurately. This function could assist in walking, running, jumping or lifting objects that are beyond the human abilities to carry. In this paper, an upper body exoskeleton robot for rehabilitation applications, called FARAT1, is presented. This exoskeleton could be used for physiotherapy of whole arm of a patient, when the physiotherapist wears the MYO armband device and performs predefined actions. So the design process of the main parts including biomechanical modeling, conceptual design aspects, loading analysis and stress analysis of the hand are presented. The manufacturing points including 3D printing of the main parts are explained and final prototype of the robot with control instruments and design mobile application for control are shown.
View Paper

Official page: IEEE Xplore

Part of the work at SYNTECH Technology & Innovation Center, this project developed a wearable exoskeleton for upper body assistance and rehabilitation, incorporating advanced mechatronics and control systems.

📅 2017 Artificial Intelligence and Robotics (IRANOPEN)

Mechanical Basic and Detailed Design for the Redundant Arm SAAM applied on a Domestic Service Robot

Authors: Farzad Cheraghpour Samavati, Majid Iranikhah, Parastoo Dastangoo, Mohamad Reza Shahabian Alashti
Published In Artificial Intelligence and Robotics (IRANOPEN)
Year 2017
Abstract
In this paper we describe design and manufacturing process of mechanical robotic arm SAAM (seven axis anthropomorphic manipulator). The main goal is designing a suitable arm for use as a service robot arm in the home environment. Design is described in the five step: Loading analysis, Stress Analysis, Material Selection, Failure Theory consideration, Safety Factor calculation. All steps calculation are done for each part of the arm. Finally, the process of manufacturing the arm is explained and the real sample of the arm is manufactured.
View Paper

Official page: IEEE Xplore

Comprehensive design work for the SAAM (7-DoF) robotic arm used in domestic service robots, covering:

  • Kinematic analysis and workspace optimization
  • Mechanical component selection and design
  • Manufacturing and assembly procedures
  • Integration with mobile robot platforms
  • Control system architecture

This arm was successfully deployed on service robots competing in RoboCup @Home league.