Mobile applications on smartphones and tablets have become part of our everyday lives. The number of augmented reality (AR) technology applications is also increasing. Augmented reality has proven to be effective in various areas of human life, from education, marketing, and training to navigation. All people have the right to access information and use available technologies, but not everyone has the same opportunities. To contribute to the digital inclusion of people who are often disadvantaged in this regard, we should think about the accessibility of digital technologies, including mobile augmented reality (MAR). The specificity of MAR is a new way of human–computer interaction compared to traditional mobile solutions. The objective of this review paper is to analyze the handheld AR solutions developed for people with different disabilities to identify accessibility challenges related to interaction when performing different tasks in AR. It also explores and presents accessibility features and other accessibility best practices, as well as potential future research directions related to the personalization and customization of such solutions for individuals. The results of this literature review can contribute to the creation of accessibility guidelines in the field of handheld AR and encourage the development of accessible AR solutions that can benefit not only people with disabilities but also people without disabilities.

In this study, we tested the ability of a machine-learning model (ML) to evaluate different user interface designs within the defined boundaries of some given software. Our approach used ML to automatically evaluate existing and new web application designs and provide developers and designers with a benchmark for choosing the most user-friendly and effective design. The model is also useful for any other software in which the user has different options to choose from or where choice depends on user knowledge, such as quizzes in e-learning. The model can rank accessible designs and evaluate the accessibility of new designs. We used an ensemble model with a custom multi-channel convolutional neural network (CNN) and an ensemble model with a standard architecture with multiple versions of down-sampled input images and compared the results. We also describe our data preparation process. The results of our research show that ML algorithms can estimate the future performance of completely new user interfaces within the given elements of user interface design, especially for color/contrast and font/layout.

In recent years, the field of artificial intelligence has experienced significant growth, which has been primarily attributed to advancements in hardware and the efficient training of deep neural networks on graphics processing units. The development of high-quality artificial intelligence solutions necessitates a strong emphasis on data-centric approaches that involve the collection, labeling and quality-assurance of data and labels. These processes, however, are labor-intensive and often demand extensive human effort. Simultaneously, there exists an abundance of untapped data that could potentially be utilized to train models capable of addressing complex problems. These raw data, nevertheless, require refinement to become suitable for machine learning training. This study concentrates on the computer vision subdomain within artificial intelligence and explores data requirements within the context of requirements engineering. Among the various data requirement activities, label quality assurance is crucial. To address this problem, we propose a machine learning-based method for automatic label quality assurance, especially in the context of object detection use cases. Our approach aims to support both annotators and computer vision project stakeholders while reducing the time and resources needed to conduct label quality assurance activities. In our experiments, we trained a neural network on a small set of labeled data and achieved an accuracy of 82% in differentiating good and bad labels on a large set of labeled data. This demonstrates the potential of our approach in automating label quality assurance.

This paper presents a systematic literature review on the use of machine learning (ML) for developing adaptive accessible user interfaces (AUI) with emphasis on applications in emerging technologies such as augmented and virtual reality (AR/VR). The review, conducted according to the PRISMA 2020 methodology, included 57 studies published between 2018 and 2025. Among them we identified 24 papers explicitly describing ML-based adaptive interface solutions. Supervised learning was dominant (83% of studies) with only isolated cases of reinforcement, generative AI, and fuzzy–NLP hybrid paradigms. The analysis of all 57 papers included in review revealed that adaptive interfaces dominate current research (65%), while intelligent or hybrid systems remain less explored. Mobile platforms were the most prevalent implementation environment (25%), followed by web-based (19%) and multi-platform systems (11%), with immersive (VR/XR) and IoT contexts still emerging. Among 43 studies addressing accessibility, the most were focused on visual impairments (33%), followed by cognitive and learning disorders (25%). The results of this review can inform the creation of accessibility guidelines in emerging AR and VR applications and support the development of inclusive solutions that benefit people with disabilities, older adults, and the general population. The main contribution of this paper lies in identifying existing gaps in the integration of accessibility and Universal Design principles into ML-based adaptive systems and in proposing a new AUI model that enables user-approved, time-delayed adaptations through machine learning, balancing autonomy, personalization, and user control.

Deep regression models are widely employed to solve computer vision tasks, such as human age or pose estimation, crowd counting, object detection, etc. Another possible area of application, which to our knowledge has not been systematically explored so far, is proportion judgment. As a prerequisite for successful decision making, individuals often have to use proportion judgment strategies, with which they estimate the magnitude of one stimulus relative to another (larger) stimulus. This makes this estimation problem interesting for the application of machine learning techniques. In regard to this, we proposed various deep regression architectures, which we tested on three original datasets of very different origin and composition. This is a novel approach, as the assumption is that the model can learn the concept of proportion without explicitly counting individual objects. With comprehensive experiments, we have demonstrated the effectiveness of the proposed models which can predict proportions on real-life datasets more reliably than human experts, considering the coefficient of determination (>0.95) and the amount of errors (MAE < 2, RMSE < 3). If there is no significant number of errors in determining the ground truth, with an appropriate size of the learning dataset, an additional reduction of MAE to 0.14 can be achieved. The used datasets will be publicly available to serve as reference data sources in similar projects.

Serious games combined with augmented reality (AR) can positively impact user’s motivation and learning experience. To ensure that users with different needs can use them effectively, we must ensure the accessibility of such solutions. The goal of this paper is to show that game-learning analytics can be utilized to identify issues for people with various needs and define ways to improve the accessibility of AR-based serious games. This paper presents a framework based on analysing automatically collected user interaction data in handheld AR-based serious games to adapt them to the user in the context of accessibility. An experiment was conducted with people with disabilities and people of different ages using the established framework. The focus of the experiment results analysis in this paper is on the evaluation of interaction mechanisms adapted to users based on their interaction data and preferences. The results showed that for most participants the assigned interaction mechanisms for virtual object transformation tasks (translation, scaling, rotation) in AR were satisfactory. The framework proposed in this paper provides the basis for further research in this field concerning the use of advanced analytics that enable intelligent adaptation to the individual user.