Advances in artificial intelligence and embodied interaction in augmented reality (AR) are creating new opportunities for intelligent instructional systems that dynamically adapt to individual learners. However, sustaining real-time responsiveness while preserving natural, socially meaningful interaction remains a persistent challenge. This work introduces ARIA (Augmented Reality Instructional Agent), a real-time software architecture for embodied AI instruction in augmented reality. ARIA leverages large language models (LLMs) with adaptive prompt engineering to tailor dialogue style, instructional strategy, and persona expression to each user. Its modular pipeline is optimized for robust, low-latency performance, with benchmarks reported for responsiveness and system stability. To complement the technical evaluation, user experience was assessed through standardized questionnaires, offering insights into perceived personalization, trust, and interaction quality. Quantitative and qualitative results demonstrate that ARIA achieves sub-second responsiveness, high pragmatic and hedonic usability, and a strong sense of co-presence and instructional trust. This work contributes a unified framework and reference architecture for developing adaptive embodied agents that combine technical efficiency with human-centered design, highlighting how real-time responsiveness can serve as the foundation for relational engagement in embodied AI instruction.

While embodied conversational agents driven by Large Language Models (LLMs) are emerging as valuable tools for instruction in Augmented Reality (AR), a key challenge lies in crafting their personalities to optimize both instructional efficacy and user engagement. To address this, we present findings from a within-subjects experiment that compared task performance and user experience with a LEGO assembly task. Participants received guidance from a real human instructor and three virtual counterparts, whose Big Five personality profiles were designed to be: (1) a direct replica of the real human, (2) an "ideal" profile ba sed on pedagogical research, or (3) customized by the participant. Our results reveal a critical trade-off: instruction from the real expert resulted in superior task efficiency and clarity; however, among the virtual conditions, instructors with idealized or user-customized personalities fostered significantly higher levels of user engagement and social presence compared to the virtual replica. Crucially, allowing users to customize their instructor's persona led to the strongest preference for future interaction. These findings underscore that personality is a fundamental component in the design of AI-driven instructors, providing empirical evidence for navigating the balance between task-oriented guidance and personalized, socially resonant user experiences.

The swift advancement of drone technology presents new challenges in data analysis and integration as deployments increase. This paper addresses the complexities of managing multiple drone video feeds and controlling autonomous drone swarms to enhance situational awareness through the use of real-time updated digital twins for drone swarm command and control. We aim to enhance the effectiveness of drone swarm operations by introducing a novel method, Camera-based Adaptive Line Formation and Dynamic Leader-following Optimization (CALF-DLFO). We investigate the impact of this novel control method on area coverage and drone distribution, considering factors such as formation direction, speed, separation distance, and adjustments. Simulation experiments demonstrate a significant improvement in area coverage compared to prior methods which focused on isolated segments. Our camera-based adaptive line formation, combined with efficient drone control mechanisms, not only enhances coverage but also provides a promising approach to scalable and automated drone swarm management.

It is not uncommon for science fiction movies to portray futuristic user interfaces that can only be realized decades later with state-of-the-art technology. In this work, we present a prototypical augmented reality (AR) installation that was inspired by the movie The Time Machine (2002). It consists of a transparent screen that acts as a window through which users can see the stereoscopic projection of a three-dimensional virtual human (VH). However, there are some key differences between the vision of this technology and the way VHs on these displays are actually perceived. In particular, the additive light model of these displays causes darker VHs to appear more transparent, while light in the physical environment further increases transparency, which may affect the way VHs are perceived, to what degree they are trusted, and the distances one maintains from them in a spatial setting. In this paper, we present a user study in which we investigate how transparency in the scope of transparent AR screens affects the perception of a VH’s appearance, social presence with the VH, and the social space around users as defined by proxemics theory. Our results indicate that appearances are comparatively robust to transparency, while social presence improves in darker physical environments, and proxemic distances to the VH largely depend on one’s distance from the screen but are not noticeably affected by transparency. Overall, our results suggest that such transparent AR screens can be an effective technology for facilitating social interactions between users and VHs in a shared physical space.

Imaging spectroscopy holds specific applications in mapping natural vegetation structures for nature conservation and ecological quantities for a better process understanding. However, common methods of vegetation classification often fail to cover variability of ecological surface parameter, especially in transition zones. In this paper we analyze the potential of two hyperspectral sensors (HyMap, AISA) to bridge ecological models with spectral information in order to map vegetational continuum, habitat types as well as assessment categories for legal authorities. As a first step a new method of aggregating ordination space structure via Kriging estimators were introduced to parameterize vegetation characteristics. With regard to future sensor design a higher spectral resolution could be detected as preferable in predicting ordination axes metrics with PLS-models. Furthermore an increasing spatial resolution was identified as a key factor observing small scale heterogeneity for habitat encroachment whereas habitat distinction can be adequately realized on the basis of occurrence probabilities.

The increased amount of software in automotive embedded systems has challenged its C code development to successfully manage software design, reuse, flexibility and efficient implementation. Model-based methods help to address such challenges with more abstract specification, code generation and simulation to determine if software design will meet requirements. However, in todaypsilas development processes lots of different legacy artifacts are involved, that hamper a frictionless migration from C code to model-based design. Therefore, migration concepts and adequate domain-specific methods with adoption of modeling languages and their tools in established embedded coding environments are needed. In our approach we present a novel migration concept considering the integration of two different modeling languages UML and Simulink in a traditional automotive software engineering process. The proof is demonstrated within the software development of a real automotive car door-controller ECU.

To support context-aware applications, it is beneficial to maintain shared context models that contain different types of information, like mobile objects, stationary objects, or spatially related digital information. This demonstration is about the NexusEditor, a graphical user interface to maintain spatial context models, interactively create queries, send them to a server and visualize the results. The contribution here is to show how schema awareness can improve such a tool: the NexusEditor dynamically parses the underlying data model and provides additional syntactic and semantic checks and short-cuts based on the schema information. Also, it supports export to existing information spaces like GoogleEarth.