Swarm fragmentation, the breakdown of communication and coordination among robots, can critically compromise a swarm's mission. Integrating Augmented Reality support into swarm monitoring-especially through co-located visualisations anchored directly on the robots- may enable human operators to detect early signs of fragmentation and intervene effectively. In this work, we propose three localised visual cues-targeting robot connectivity, dominant decision influences, and movement direction-to make explicit the underlying Perception-Decision-Action (PDA) loop of each robot. Through an immersive Virtual Reality user study, 51 participants were tasked with both anticipating potential fragmentation and selecting the appropriate control to prevent it, while observing swarms exhibiting expansion, densification, flocking, and swarming behaviours. Our results reveal that a visualisation emphasising inter-robot connectivity significantly improves anticipation of fragmentation, though none of the cues consistently enhance control selection over a baseline condition. These findings underscore the potential of co-located AR-enhanced visual feedback to support human-swarm interaction and inform the design of future AR-based supervisory systems for robot swarms. A free copy of this paper and all supplemental materials are available at https://osf.io/49gny.

Attentional tunneling, a phenomenon where operators focus excessively on one task or channel of information while neglecting others, poses significant risks in critical, multitasking environments such as aviation, nuclear power, and cybersecurity. This study explores the use of Augmented Reality (AR) to mitigate attentional tunneling and enhance task performance by redirecting attention effectively across multiple visual cues. A user experiment involving eighteen participants was conducted to evaluate the effectiveness of two types of AR cues-Minimap and Line-compared to a control condition with no AR assistance. Participants performed a series of tasks using a head-mounted display (HMD) while interacting with a touchscreen in a simulated environment. Results show that both AR cues significantly reduced missed alerts and decreased cognitive workload, with the Line cue proving slightly more effective in reducing response time to peripheral alerts. The findings suggest that AR-based interventions can improve attention management and task performance in complex systems by countering the effects of attentional tunneling. This study highlights the potential of AR technology to enhance operational safety and efficiency in high-stakes environments.

As artificial intelligence (AI) is increasingly present in different aspects of society and its harmful impacts are more visible, concrete methods to help design ethical AI systems and limit currently encountered risks must be developed. Taking the example of a well-known Operations Research problem, the Nurse Rostering Problem (NRP), this paper presents a way to help close the gap between abstract principles and on-the-ground applications with two different steps. We first propose a normative step that uses dedicated scientific knowledge to provide new rules for an NRP model, with the aim of improving nurses’ well-being. However, this step alone may be insufficient to comprehensively deal with all key ethical issues, particularly autonomy and explicability. Therefore, as a complementary second step, we introduce an interactive process that integrates a human decision-maker in the loop and allows practical ethics to be applied. Using input from stakeholders to enrich a mathematical model may help compensate for flaws in automated tools.

Interactive systems based on Artificial Intelligence (AI) algorithms are raising new challenges, including establishing a bond of trust between users and AI. This trust must be calibrated to match the degree of reliability of AI in order to avoid over-trusting and under-trusting. However, trust is a subjective characteristic that is difficult to assess as it can vary from one person to another. This paper explores how it is possible to estimate the trust of users, especially through behavioral and physiological sensing. It also explains how, from trust assessment, it becomes possible to develop techniques for calibrating trust.

Rationale: The advent of Industry 5.0 places a heightened focus on enhancing worker wellbeing during the digital transformation of factories. System models that ignore human workers yield suboptimal results in product design and system improvement.Purpose: In the aircraft industry, worker workload is of primary concern as most tasks are performed manually, leading to general fatigue and musculoskeletal disorders. Robot assistance could improve quality, efficiency and relieve workers from fatigue. To demonstrate the feasibility and value of integrating human performance models in system design at Airbus, a Worker Fatigue Model was developed, focusing on the effects of (1) automation (manual vs semi-automated), and (2) workforce makeup (various ratios of high-skilled to low-skilled workers). Our ultimate goal was to inform the development of effective policies and strategies for human-technology integration in Industry 5.0.Methods: We developed the Worker Fatigue Model by adapting existing fatigue models for workers in industrial environments and by considering worker characteristics, tasks, and the presence or absence of robot-assistance. Two different scenarios were simulated (fully manual and semi-automated), with input variables such as worker skill, age, and motivation, and output variables including overall fatigue and error probabilities were evaluated. The Worker Fatigue Model was integrated into the Airbus system model to conduct trade studies based on workforce characteristics.Results: Our findings revealed that the composition of the workforce (i.e., various ratios of high-skilled to low-skilled workers), alongside specific manufacturing technologies, significantly reduced worker fatigue, especially with higher ratios of high-skilled workers, and improved overall industrial system performance.Conclusions: Although applying our Worker Fatigue Model effectively demonstrated the feasibility and value of integrating human factors into early industrial system design, it remains a work in progress. Future work will aim to accurately represent the workload of human workers, including operational costs, when implementing robot assistance.

Pour répondre aux problèmes posés par l'utilisation croissante des modèles IA dans les applications à forts enjeux socio-économiques ou de sécurité, l'intelligence artificielle explicable (XAI) a connu un essor important durant les dernières années.Initialement dévolue à la recherche de solutions techniques permettant de produire automatiquement des explications, elle s'est heurtée à plusieurs difficultés, en particulier lorsque ces solutions ont été confrontées à des utilisateurs finaux non experts.L'XAI s'est alors attachée à s'inspirer des sciences sociales pour produire des explications plus faciles à comprendre.Malgré certains résultats encourageants, cette nouvelle approche n'a pas apporté autant qu'espéré.Cet article analyse l'évolution de l'XAI à travers ces deux périodes.Il évoque des raisons possibles des difficultés rencontrées, puis propose une nouvelle approche pour améliorer la production automatisée d'explications.Cette approche, nommée explicabilité sémantique ou S-XAI, est centrée sur la cognition de l'utilisateur.Alors que les méthodes précédentes sont orientées sur les algorithmes ou sur la causalité, la S-XAI part du principe que la compréhension repose avant tout sur la capacité de ce dernier à s'approprier le sens de ce qui est expliqué.ABSTRACT.To respond to the problems posed by the growing use of AI models in high stakes applications, explainable artificial intelligence (XAI) has experienced significant growth in recent years.Initially dedicated to the search for technical solutions making it possible to produce explanations automatically, it encountered several difficulties, in particular when these solutions were confronted with non-expert end users.The XAI then sought to draw inspiration from the social sciences to produce explanations that were easier to understand.Despite some encouraging results, this new approach has not brought as much as hoped.This article analyzes the evolution of the XAI through these two periods.He discusses possible reasons for the difficulties encountered, and then proposes a new approach to improve the automated production of explanations.This approach, called semantic explainability or S-XAI, focuses on user cognition.While previous methods are oriented towards algorithms or causality, S-XAI starts from the principle that understanding relies above all on the user's ability to appropriate the meaning of what is explained.

Swarms of Unmanned Combat Aerial Vehicles (UCAVs) are efficient in various tasks. However, they evolve in hostile environments with risks of destruction during their flight. To mitigate this risk, it is known that cooperative behaviour can be used to enhance the protection within the swarm. The goal of this paper is to design efficient algorithms to guide the overall swarm to a given target while minimizing the risk of destruction of the member of the swarm. First, a new model, based on a controlled Markov chain, is derived to capture this cooperative swarm effect on the destruction threat of each member of the swarm. Then, an algorithm combining path planning to guide the overall swarm and local individual control to optimize the formation is suggested to help a swarm to reach a target before the destruction of all UCAVs. We evaluate our approach using numerical experiments.

The LOTUS project aims at improving maritime surveillance. In this context, this position paper presents ongoing contributions, including novel machine learning algorithms for multi-agent systems to be applied to groups of underwater drones involved in surveillance missions. It emphasises incorporating human-machine teaming to bolster decision-making in maritime scenarios. The expected outcomes of this project comprise the robust control of groups of autonomous vehicles, adaptable to environmental changes, as well as an effective reporting method. Mission summaries will be delivered to human operators by way of narratives about the relevant events detected thanks to drones. The integration of this narrative construction powered by machine learning will enhance the overall effectiveness of the team, constituting a significant breakthrough.

In the context of robot swarms, fragmentation refers to a breakdown in communication and coordination among the robots. This fragmentation can lead to issues in the swarm self-organisation, especially the loss of efficiency or an inability to perform their tasks. Human operators influencing the swarm could prevent fragmentation. To help them in this task, it is necessary to study the ability of humans to perceive and anticipate fragmentation. This article studies the perception of different types of fragmentation occurring in swarms depending on their behaviour selected amongst swarming, flocking, expansion and densification. Thus, we characterise human perception thanks to two metrics based on the distance separating fragmented groups and the separation speed. The experimentation protocol consists of a binary discrimination task in which participants have to assess the presence of fragmentation. The results show that detecting fragmentation for expansion behaviour and anticipating fragmentation, in general, are challenging. Moreover, they show that humans rely on separation distance and speed to infer the presence or absence of fragmentation. Our study paves the way for new research that will provide information to humans to better anticipate and efficiently prevent the occurrence of swarm fragmentation.

Self-organisation in robot swarms can produce collective behaviours, particularly through spatial self-organisation. For example, it can be used to ensure that the robots in a swarm move collectively. However, from a designer’s point of view, understanding precisely what happens in a swarm that allows these behaviours to emerge at the macroscopic level remains a difficult task. The same behaviour can come from multiple different controllers (ie the control algorithm of a robot) and a single controller can give rise to multiple different behaviours, sometimes caused by slight changes in self-organisation. To grasp the causes of these differences, it is necessary to investigate the relationships between the many methods of self-organisation that exist and the various behaviours that can be obtained. The work presented here addresses self-organisation in robot swarms by focusing on the main behaviours that lead to spatial self-organisation of the robots. First, we propose a unified definition of the different behaviours and present an original classification system highlighting ten self-organisation methods that each allow one or more behaviours to be performed. An analysis, based on this classification system, links the identified mechanisms with behaviours that could be considered as obtainable or not. Finally, we discuss some perspectives on this work, notably from the point of view of an operator or designer.

Real-time and high-intensity teamwork management is complex, as team leaders must ensure good results while also considering the well-being of team members. Given that stress and other factors directly impact team members’ output volume and error rate, these team leaders must be aware of and manage team stress levels in combination with allocating new work. This paper examines methods for visualizing each team member’s status in mixed reality, which, combined with a simulated stress model for virtual team members, allows the team leader to take into account team members’ individual statuses when choosing whom to allocate work. Using simulated Augmented Reality in Virtual Reality, a user study was conducted where participants acted as team leaders, putting simulated team members under stress by allocating several required work tasks while also being able to review the stress and status of each team member. The results showed that providing Augmented Reality feedback on team members’ internal status increases the team’s overall performance, as team leaders can better allocate new work to reduce team members’ stress-related errors while maximizing output. Participants preferred having a graph representation for stress levels despite performing better with a text representation.

Collaborative virtual environments allow people to work together while being distant. At the same time, empathic computing aims to create a deeper shared understanding between people. In this paper, we investigate how to improve the perception of distant collaborative activities in a virtual environment by sharing users’ activity. We first propose several visualization techniques for sharing the activity of multiple users. We selected one of these techniques for a pilot study and evaluated its benefits in a controlled experiment using a virtual reality adaptation of the NASA MATB-II (Multi-Attribute Task Battery). Results show (1) that instantaneous indicators of users’ activity are preferred to indicators that continuously display the progress of a task, and (2) that participants are more confident in their ability to detect users needing help when using activity indicators.

To improve the safety and the performance of operators involved in risky and demanding missions (like drone operators), human-machine cooperation should be dynamically adapted, in terms of dialogue or function allocation. To support this reconfigurable cooperation, a crucial point is to assess online the operator's ability to keep performing the mission. The article explores the concept of Operator Functional State (OFS), then it proposes to operationalize this concept (combining context and physiological indicators) on the specific activity of drone swarm monitoring, carried out by 22 participants on simulator SUSIE. With the aid of supervised learning methods (Support Vector Machine, k-Nearest Neighbors, and Random Forest), physiological and contextual are classified into three classes, corresponding to different levels of OFS. This classification would help for adapting the countermeasures to the situation faced by operators.

This paper addresses the challenge of embedded computing resources required by future autonomous Unmanned Aircraft Systems (UAS). Based on an analysis of the required onboard functions that will lead to higher levels of autonomy, we look at most common UAS tasks to first propose a classification of UAS tasks considering categories such as flight, navigation, safety, mission and executing entities such as human, offline machine, embedded system. We then analyse how a given combination of tasks can lead to higher levels of autonomy by defining an autonomy level. We link UAS applications, the tasks required by those applications, the autonomy level and the implications on computing resources to achieve that autonomy level. We provide insights on how to define a given autonomy level for a given application based on a number of tasks. Our study relies on the state-of-the-art hardware and software implementations of the most common tasks currently used by UAS, also expected tasks according to the nature of their future missions. We conclude that current computing architectures are unlikely to meet the autonomy requirements of future UAS. Our proposed approach is based on dynamically reconfigurable hardware that offers benefits in computational performance and energy usage. We believe that UAS designers must now consider the embedded system as a masterpiece of the system.

This paper addresses the question of online fatigue monitoring in high constrained work environments, by dealing more specifically with the activity of submariners. A state of the science is proposed on the concept of fatigue as well as physiological and behaviour metrics supporting the emergence of a fatigue management system for individuals and teams. From this, a framework for online fatigue monitoring in maritime environments is proposed.

To improve the safety and the performance of operators involved in risky and demanding missions, human-machine cooperation should be dynamically adapted, in terms of dialogue or function allocation. To support this reconfigurable cooperation, a crucial point is to assess online the operator’s ability to keep performing the mission, to anticipate and predict potential future performance impairments, as well as to be able to activate appropriate countermeasures in time. Thus, the paper explores the concept of Operator Functional State (OFS) developed by Hockey in 2003, by articulating it with underlying cognitive and attentional states, as well as with the notion of cognitive control modes.