Papers & articles

2025

Long Short-Term Memory-Enabled Electromyography-Controlled Adaptive Wearable Robotic Exoskeleton for Upper Arm Rehabilitation

Biomimetics — MDPI

A wearable robotic exoskeleton powered by EMG data and LSTM neural networks achieves highly accurate upper arm rehabilitation with 96% prediction accuracy, 89% exoskeleton control accuracy, and 97% range of motion correction performance, offering personalised adaptive support for patients recovering from illness, accidents, or surgery.

Read article doi.org/10.3390/biomimetics10020106

2026

Intelligent Ergonomic Chair with AI-Driven Posture Analysis and Correction with Real-Time Feedback

2026 Moratuwa Engineering Research Conference (MERCon)

This publication presents an intelligent ergonomic chair that reduces musculoskeletal risks from prolonged sitting by combining pressure sensing, deep learning posture recognition, automated pneumatic correction, and generative AI guidance in a unified IoT platform. A 10-channel FSR array and ESP32 capture seated pressure data, while an MLP model classifies eight postures with 99.5% accuracy on unseen participants. Sustained poor posture triggers a closed-loop pneumatic lumbar mechanism to improve spinal alignment automatically. A real-time dashboard integrated with Google Gemini provides personalized ergonomic recommendations. A 30-participant usability study demonstrated an 82.0% reduction in sustained poor posture (p < 0.001).

Read article doi.org/10.1109/MERCon71835.2026.11691539

2026

AI-Powered Braille Recognition: A YOLOv8 and LLM Approach to Real-Time Multilingual Speech Conversion

2026 IEEE International Research Conference on Smart Computing and Systems Engineering (SCSE)

This paper presents a portable, real-time Braille-to-speech conversion system that combines YOLOv8-based Braille recognition with LLM-assisted text refinement and multilingual text-to-speech to improve information accessibility for visually impaired users. Implemented on a Raspberry Pi 5, the system achieves high recognition accuracy and real-time processing performance, providing an affordable and scalable assistive solution for multilingual environments.

Read article doi.org/10.1109/SCSE70081.2026.11499975

2026

A Deep Bi-LSTM-Based Motor Imagery Brain–Computer Interface for Wheelchair Navigation

2026 IEEE International Research Conference on Smart Computing and Systems Engineering (SCSE)

This paper presents a Brain–Computer Interface (BCI)-based wheelchair navigation system that uses a deep Bi-LSTM network to classify motor imagery EEG signals into navigation commands with high accuracy and reliability. Integrated with a cloud-assisted architecture and ESP32-based embedded control, the system enables responsive, robust and practical assistive mobility for individuals with severe motor impairments.

Read article doi.org/10.1109/SCSE70081.2026.11499971

2026

LSTM-Based Fault Diagnosis for Industrial Blast Freezers Using Multisensor Time-Series Data

2025 IEEE 19th International Conference on Industrial and Information Systems (ICIIS)

This paper presents an IoT-based predictive maintenance system for commercial blast freezers that combines custom ESP32 sensor networks with LSTM deep learning to monitor current, vibration and temperature data in real time. The system achieves high prediction accuracy, outperforming conventional machine learning models, while enabling proactive maintenance through real-time monitoring and predictive alerts to reduce equipment downtime.

Read article doi.org/10.1109/ICIIS69028.2026.11450878

2026

Integrating Adaptive Cruise Control into Autonomous Wheelchair Navigation for Unstructured Outdoor Terrain

2025 IEEE 19th International Conference on Industrial and Information Systems (ICIIS)

This paper presents ALEX, an adaptive cruise control framework for autonomous wheelchairs that combines YOLOv8-based scene segmentation, occupancy grid mapping and potential-field path planning to enable safe navigation in unstructured outdoor environments. The system achieves real-time performance with high detection and navigation accuracy, demonstrating reliable, infrastructure-free mobility that enhances the independence of users with severe lower limb impairments.

Read article doi.org/10.1109/ICIIS69028.2026.11450820

2026

Environment-Aware Intention Generation in Robots via Multistage Perception and Language Reasoning

2025 IEEE 19th International Conference on Industrial and Information Systems (ICIIS)

This paper presents a real-time humanoid intent generation framework that combines YOLOv8x-based object detection, ANN-based environment classification and a hybrid LLM reasoning mechanism to enable context-aware, human-like behaviour in service robots. Deployed on a cloud-edge architecture, the system achieves high perception accuracy, low-latency inference and adaptive intent generation, demonstrating its effectiveness for real-time assistive and socially intelligent robotic applications.

Read article doi.org/10.1109/ICIIS69028.2026.11450615

2025

Vision-Driven Diagnostic Intelligence Through Deep Learning for Retinal Health Analytics

2025 Moratuwa Engineering Research Conference (MERCon)

This study uses retinal fundus images with ResNet101 and U-Net models to classify ARMD, DR and key retinal lesions with high precision and segmentation accuracy. The results demonstrate strong potential for automated vascular health monitoring using deep learning.

Read article doi.org/10.1109/MERCon67903.2025.11217004

2025

Sinhala Language Specific Vocal Biomarker Extraction for Parkinson's Diagnosis Using Machine Learning

2025 Moratuwa Engineering Research Conference (MERCon)

Early Parkinson's detection was explored using Sinhala voice biomarkers by analysing vowel sounds from Sri Lankan patients with machine learning models. A hybrid SVM–Logistic Regression classifier achieved 87% accuracy, showing strong potential for low-resource screening.

Read article doi.org/10.1109/MERCon67903.2025.11217005

2024

Assistive Cartesian Healthcare Robot for Forearm Vein Puncturing

2024 6th International Conference on Advancements in Computing (ICAC)

A medical robot using Cartesian coordinate mechanisms and advanced sensor technology achieves highly accurate automated venipuncture with 90.9% box prediction accuracy, 94.9% vein segmentation accuracy and a 98.7% puncturing success rate, reducing human error and improving patient outcomes in healthcare settings.

Read article doi.org/10.1109/ICAC64487.2024.10851158