In this paper we present PuReWidgets, a toolkit that supports multiple interaction mechanisms, asynchronous events, automatic web and mobile interface generation, and concurrent interaction. This is an early effort towards the creation of a programming toolkit that developers can incorporate into their public display applications to support the interaction process across multiple display systems without considering the specifics of what interaction modality will be used on each particular display.

This thesis systematically investigates strategies for real-time voluntary muscle contrac- tion detection and command recognition in an electromyography (EMG)-based human- computer interface (HCI), aiming to improve adaptability, robustness, and usability for individuals with motor neuron diseases, particularly Amyotrophic Lateral Sclerosis (ALS). Building on a previously developed EMG-based interface combining convolutional neural networks with long short-term memory layers for EMG classification, this work evaluates this architecture performance under different conditions and compares it with simpler threshold-based models. The study investigates the effects of preprocessing, dataset composition, calibration strategies, and recognition criteria across three types of datasets collected from both able- bodied and ALS participants during an interaction task. Results show that while standard signal conditioning, including DC offset removal, band-pass filtering, rectification, and envelope extraction, stabilizes EMG signals and reduces noise, it slightly decreases model performance by limiting feature richness, with normalization further introducing cali- bration dependencies and reducing precision. Including ALS data in training enhanced generalisation to pathological signals, and calibration based solely on rest proved feasible, reducing user effort. A percentile-based energy threshold was proposed as a foundation for dynamic, user-adaptive calibration. Finally, adjusting the rules for distinguishing between short and long clicks significantly improved recognition performance, aligning system output with user actions. These findings refine a functional prototype and establish a systematic modular frame- work for developing more reliable, adaptive, and clinically meaningful solutions for ALS patients. This work supports the development of next-generation HCIs that enhance communication and autonomy by integrating emerging AI models and incorporating data from individuals with ALS.

Nowadays more than 700 million persons around the world have some kind of disability or handicap. During the last decades the elderly population in most of the European countries and across all the most civilized countries is also growing at an increasing pace. This phenomenon is receiving increasing attention from the scientific community, during the last years, and several solutions are being proposed in order to allow a more independent life to the people belonging to those groups. In this context Intelligent Wheelchairs (IW) are instruments that are a natural development of the scientific work that has been conducted to improve the traditional Wheelchair characteristics using health informatics, assistive robotics and human computer interface technologies. Some of the most important features of the IW are their navigation capabilities and automatic adaptation of their interface to the user. This paper presents the evolution and state of art concerning IWs prototypes and simulators and intelligent human-computer interfaces in the context of this devices. Our study enabled us to conclude that although several Intelligent Wheelchair prototypes are being developed in a large number of research projects, around the world, the adaptation of their user interface to the patient is an often neglected research topic. Thus, projects aiming at developing new concepts of Intelligent Wheelchairs are needed mainly using multimodal interfaces and wheelchair interfaces adapted to the user characteristics.