Open research questions in Human-Automation Interaction and Safety

What the literature leaves open

• In this study, it was investigated whether the outlined collection of empirical methods is suitable to analyze visual, mixed-reality assistance systems in an air-to-air-refueling simulation with the drogue-and-probe method and if it can highlight differences regarding HF aspects such as WL, SA, usability and UX. The findings of the study indicate, that the applied methodology can objectify the comparison of different visual, mixed-reality assistance systems and can draw attention to differences in terms of HF aspects and pilot behavior. It can be concluded that hypothesis H2, which addresses the applicability of the selected methodology for evaluating visual, mixed-reality assistance systems in the refueling process using the probe-and-drogue method, can be considered fulfilled. However, not all of the methods applied and metrics used were able to deliver robust results. It is assumed that this is partially due to the small number of pilots tested. Therefore, and due to the limited availability of pilots for testing purposes, further investigations are needed to improve the methodology for small sample sizes. Additionally, standardizing the procedures and methods could lead to improved results. Similarly, a detailed guideline should be developed that explains how recorded simulation parameters can be used to obtain knowledge about changes in pilot behavior. Furthermore, it needs to be mentioned that multiple valuable insights into issues and potential improvements of the tested visual assistance systems emerged from unscripted conversations and questions with the pilots during the experiment, particularly during the de-briefing sessions. This emphasizes the importance of “normal”, non-scripted, subject-related conversations with highly specialized experts. Such discussions can highlight important details that may not be captured by standardized questionnaires. Consequently, it is strongly recommended to incorporate non-directive interview slots into similar experiments. Finally, it should be mentioned, that physiological measurements could be a valuable add-on to the data obtained with questionnaires and flight data recording. Especially, if physiological reactions and measurements could resolve the impacts of dedicated events during the AAR scenario and could be synchronized with the flight data, this could support the development process. In particular, the objective is to incorporate eye-tracking in the forthcoming assessments to gain a deeper understanding of pilot engagement with the presented assistance systems and their effect on WL. The knowledge gained from this study and the unresolved questions will be further analyzed in the DLR e.V.

  Human factors evaluation of mixed reality assistance systems for aerial refueling of fighter aircraft · 2026 · DOI

• The findings are limited to simulation and are intended to provide preliminary insights into the technical and safety implications of incorporating AI-based radar perception within ISO 26262-compliant ADAS architectures.

  Harmonizing Generative AI and ISO 26262 for Safe and Intelligent ADAS Systems · 2026 · DOI

• Therefore, future research should examine how perceived presence, environmental perception, ECL, and OCL jointly explain learning outcome in IVR-based warehouse safety training.

  Comparing the predictive roles of intrinsic and extraneous cognitive load on learning performance: the mediating effect of IVR self-efficacy in warehouse training · 2026 · DOI

• Papenfuss and Schmidt, 2023) and the interplay of team task and team communication, empirical evidence of stable communication pattern which could be observed in several studies are scarce.

  JARVIS D5.2 Whitepaper on Human-AI teaming · 2026 · DOI

• Reference C establishes a relational vocabulary for Accessibility Geometry, but substantial work remains. Future work should proceed along eight tracks. 20.1 Relational Proxy Libraries Each substrate requires its own proxy library. Possible future proxy libraries may include: trust proxies; shared geometry proxies; relational impedance proxies; • • • • observer coupling proxies; • perceived observer value proxies; stewardship responsibility proxies; • information acquisition cost proxies; • • geometric cohesion proxies. These libraries should distinguish between measured, reconstructed, inferred, modeled, narrative, and speculative evidence. 20.2 Observer Coupling Graphs Future work may represent observer-systems as relational graphs. In such models: K_ij(t) may represent coupling strength. eta_ij(t) may represent impedance or distortion across an edge. Tr_ij(t) may represent trust between observers. C_info,ij(t) may represent information acquisition cost across a relational pathway. Such graph models may support future simulation of relational collapse, repair, synchronization, isolation, and cross-scale coordination. 20.3 Relational Human TAWS Development Reference B introduced preliminary Human TAWS logic. Reference C suggests that future Human TAWS development should include relational conditions. Possible relational Human TAWS indicators include: • declining trust; increasing relational impedance; • rising information acquisition cost; • • decreasing perceived observer value; • weakening observer coupling; collapsing shared geometry; • stewardship failure; • false relational coherence; • relational isolation drift. • These indicators should be treated as provisional until calibrated. 20.4 Cross-Scale Stewardship Models Future work should examine how higher-scale observer-systems affect the accessibility landscapes of lower-scale observers. Candidate domains include: families; schools; • • • military systems; • organizations; • • • platforms; • • governance networks. courts; institutions; synthetic-agent deployments; The goal is not to impose a universal stewardship doctrine. The goal is to identify conditions under which higher-scale observers preserve or degrade lower- scale accessibility, recovery margin, constraint preservation, and trust. 20.5 Relational Repair Modeling Future work should develop models for relational repair trajectories. Candidate repair patterns include: • Trust Rebuilding; • Shared Geometry Restoration; • Observer Re-Recognition; • Healthy Decoupling; • Stewardship Restoration; • Relational Impedance Reduction; • Relational Re-Coupling; • Corrective Dissent Integration. These should be tested against real or simulated cases to determine whether they meaningfully distinguish recovery from surface stabilization. 20.6 Integration with Reference D Reference D should formalize application declaration and calibration discipline for all Accessibility Geometry applications. For Reference C variables specifically, Reference D should require: substrate declaration; • • observer-scale declaration; • proxy declaration; • • • non-claim declaration; falsification condition. • recovery-boundary declaration; evidence-classification declaration; This will prevent relational variables from drifting into moral, legal, clinical, or metaphysical overreach. 20.7 Simulation and Toybox Investigation Future toybox investigations should simulate relational dynamics such as: trust collapse; shared geometry fracture; coercive coupling; relational isolation drift; • • • • • observer erasure; stewardship failure; • relational impedance spiral; • trust rebuilding; • • healthy decoupling; • corrective dissent integration. These simulations should distinguish relational collapse from ordinary conflict, boundary- setting, privacy protection, and healthy disagreement. 20.8 Empirical Calibration Long-term development should identify empirical or reconstructable indicators for relational variables within specific substrates. Calibration should proceed domain by domain. No relational threshold should be treated as universal. A trust proxy in a family system may not transfer to an institution, synthetic system, legal process, military unit, or online platform. The same notation may be shared. The operational meaning must remain substrate-specific.

  Accessibility Geometry Reference Stack: Reference C - Relational and Coupling Variable Register · 2026 · DOI

• 本アーキテクチャは強力な統治モデルを提供するが、同時に以下の制約事項を伴う。 • Observability limitation: Black‑box API mode や軽量ローカルモデルでは、真の hidden‑state trajectory は直接観測できない。 • Embedding dependence: Semantic drift や goal alignment は embedding quality に依存する。 • Threshold sensitivity: 重みや閾値は初期値であり、ReplayLog に基づく calibration が必要である。 • Sampling cost and latency overhead: 候補集合や Semantic Ellipsoid を扱う場合、生成回数、埋め込み 計算、スコアリング処理に追加コストが発生する。 • False sense of security: 高い convergence score は factual correctness や operational safety を保証し ない。 • Adversarial inputs: 敵対的入力は goal summaries、structured output、candidate generation を操作 し得るため、capability restriction や sandboxing と併用する必要がある。 14. 体系化された AIKernel 理論：セマンティック・コンテキスト OS 再構成された AIKernel 理論は、「セマンティック・カプセル化」、「宣言的ガバナンス」、「リソースの抽象化 と流動性」の3本柱に集約される。 Trajectory Governance は、このうち宣言的ガバナンスを担う中核要素である。これは、AIKernel をセマンティ ック OS として構成する際に、確率的推論を検証可能な実行遷移へ接続するための Governance Layer として機能 する。 15.

  AIKernel Phase-1 Paper 04: Trajectory Governance Model · 2026 · DOI

• Discussion The discussion is structured around the four research questions outlined in Section 1, linking the empirical findings to each question in turn to provide a clearer interpretation of the results. This study provides empirically grounded, trial-phase evidence from a real-world UK deployment, offering insights that extend beyond conventional perception studies based on hypothetical scenarios. Rather than demonstrating whether AS are viewed positively in general, which has been shown in elsewhere, identifying which aspects of the experience dominate users’ evaluations during early exposure, and where remaining points of friction persist even in a controlled environment.

  User experience, safety perceptions, and adoption intentions for autonomous shuttles: findings from a UK trial-phase survey · 2026 · DOI

• Future studies should consider refining the phrasing of dynamic norms to explicitly state the extent and direction of change, ensuring a clearer distinction between the two normative framings. Future studies should consider controlling for this potential confounding variable by including similar reflective tasks across all experimental conditions. after measuring A notable limitation of this study lies in the operationalization of the static and dynamic norms. While the measures had a good reliability, some should be explored further, and validated to assure their precision.

  Driven by trust? Social norms increase trust in shared autonomous public transport · 2026 · DOI

• Future research could examine SHAP values, or counterfactuals for additional perspectives on global/local. 9489) was limited to approximately 2–3 percentage points in explained vari- ance.

  Human AI trust modeling in cognitive systems via ensemble learning and advanced feature engineering · 2026 · DOI

• In this work, we present MRDrive, an open-source MR driving simulator and flexible, high-fidelity platform for prototyping and evaluating IVIS and driver interactions. By adopting MR, MRDrive combines the immersive 360°virtual driving environment with a real-world cabin mock-up, preserving haptic feedback and embodied interaction while maintaining experimental control over traffic scenarios and interface designs. The MR paradigm offers a middle ground between VR simulators (which sacrifice physical realism) and expensive high-end projection-based setups. Although it entails higher setup complexity and hardware demands than basic VR configurations, its modular open-source architecture built on Unity lowers the barrier for adoption. Through a pilot study, we demonstrated how MRDrive supports the collection and analysis of eye-tracking and touch interaction data. We see MRDrive as a reusable research instrument that reduces the barrier to studying attention, explainability, and interaction in realistic driving scenarios. However, we acknowledge that certain limitations remain. Although the Varjo XR-3 is comparatively inexpensive compared to projection-based systems, it is a high-end headset that is required to align real and virtual content with a precision of <1 cm. Additionally, alignment still requires a considerable amount of trial and error. Furthermore, in order to achieve frame rates of 40–60 fps, we use a workstation equipped with a high-end NVIDIA GeForce RTX 4090 graphics card. In addition to technical improvements such as more efficient rendering and even better alignment, future work could involve expanding the range of supported interaction modalities (e.g., gesture and voice input), integrating additional physiological sensors, and conducting further validation of the simulator against real-world driving benchmarks. This would strengthen its role in advancing human-centred design for future (automated) mobility.

  MRDrive: An Open Source Mixed Reality Driving Simulator for Automotive User Research · 2026 · DOI

• This study employed a cross-sectional design, which captures a single point in time. Consequently, while the structural equation model suggests a flow from technology anxiety through mediators to acceptance, we cannot definitively establish temporal precedence or absolute causality between these variables. Future longitudinal research would be beneficial to observe how these relationships evolve as autonomous vehicle (AV) technology matures. A notable limitation concerns the constructs derived from diffusion theory, specifically External Locus of Control (M5) and Sensation Seeking (M6). These variables exhibited low internal consistency (Cronbach’s alpha ≈ 0.50) and did not meet the recommended threshold for convergent validity (AVE < 0.50). These results may stem from the translation of original scales into Polish or semantic overlap between items. Furthermore, these two constructs did not significantly predict AV acceptance in the final model, suggesting that the current measurement of these traits may not fully capture their influence in the context of Polish AV adoption. ARTICLE IN PRESS ARTICLE IN PRESS ACCEPTED MANUSCRIPT The data was collected in Poland, a market where real-world exposure to high-level autonomous vehicles remains virtually non-existent for the general public. Responses are therefore based on hypothetical scenarios and general perceptions rather than direct interaction with the technology. This lack of experiential knowledge may influence how constructs like Trialability (M2) and Observability (M3) were interpreted by participants. As with most research relying on the survey-based data, the study is susceptible to self-report bias. While Harman’s single-factor test indicated that common method bias was not a dominant concern (explaining 35.8% of variance), the inherent limitations of this procedure mean that some degree of bias cannot be entirely ruled out. Although Age was included as a covariate and showed a small but significant negative effect on acceptance (β = -0.06, p = 0.010), this study did not perform more granular multi-group analyses. Future studies should explore whether the structural paths—particularly the mediation of technology anxiety—differ significantly across gender, socio-economic status, or levels of prior technological experience. ETHICAL CONSIDERATIONS AND INFORMED CONSENT The study was conducted in strict accordance with the ethical principles outlined in the Declaration of Helsinki. The research protocol was formally reviewed and approved by the Ethics Committee of Łukasiewicz Research Network — ITECH Institute of Innovation and Technology (Approval No. 01/12/2025). All participants were fully informed about the study’s objectives, the scope of their involvement, and the measures taken to ensure data anonymization and confidentiality. Prior to the commencement of the interviews, each respondent provided voluntary informed consent. Participants were also notified of their right to withdraw from the study at any stage without providing a justification. All personal data were handled in compliance with applicable data protection regulations (GDPR). DATA AVAILABILITY STATEMENT The data supporting the findings of this study are not publicly available to protect the anonymity of the respondents and comply with data protection regulations. However, the data are available from the corresponding author upon reasonable request. COMPETING INTERESTS STATEMENT ARTICLE IN PRESS ARTICLE IN PRESS ACCEPTED MANUSCRIPT The authors declare no conflict of interest regarding the publication of this manuscript. AUTHOR CONTRIBUTION STATEMENT M.W.: Conceptualization, Writing – original draft, Writing – review & editing; P.D. and E.K.-J.: Statistical analysis, Writing – original draft, Writing – review & editing; D.Z.: Literature review, Writing – review & editing; M.M., P.Sz., M.P., and K.Z.: Writing – review & editing. FUNDING DECLARATION The research was carried out as part of the project "Research on the perception system of autonomous vehicles in the aspect of road safety at road intersections," implemented by the Łukasiewicz Research Network – Automotive Industry Institute. This publication was co-financed from the state budget under the "Science for the Society II" program of the Minister of number NdS- II/SN/0352/2024/01, funding amount PLN 1,000,000.00, total project value PLN 1,000,000.00.

  Does fear of technology shape social acceptance of autonomous vehicles in Poland? conclusions from cross-sectional study using structural equation modelling · 2026 · DOI

• The paper does not address how the bi-directional feedback loop between the blueprint and UOMAMS performs under conditions of rapid organizational change or crisis situations.

  Of (AI) Machine and Human (Labour): An Integrated Nexus of Work Operating System Architecture for Orchestrating Human–AI Collaborations · 2026 · DOI

• There is insufficient discussion of how the framework handles edge cases or exceptions when atomic tasks do not fit neatly into the AAA (Automate/Augment/Assist) categorization scheme.

  Of (AI) Machine and Human (Labour): An Integrated Nexus of Work Operating System Architecture for Orchestrating Human–AI Collaborations · 2026 · DOI

• The paper does not explicitly discuss how the UOMAMS framework scales across large, geographically distributed organizations with thousands of atomic tasks and multiple AI agents operating simultaneously.

  Of (AI) Machine and Human (Labour): An Integrated Nexus of Work Operating System Architecture for Orchestrating Human–AI Collaborations · 2026 · DOI

• The ability of users to better interpret explainable AI influences their opinions on fairness and reliability, primarily in high-value operational environments such as aviation maintenance, continuing care, and industrial engineering; further research is needed on explainability mechanisms.

  Artificial intelligence system reliability and knowledge identity: A model for knowledge workers in knowledge management environments · 2026 · DOI

• The research was undertaken within a solitary organisation in aviation maintenance; future research should seek to explore multi-organisations with case sampling from a range of industries and levels of AI maturity to improve the generalisability of empirical work.

  Artificial intelligence system reliability and knowledge identity: A model for knowledge workers in knowledge management environments · 2026 · DOI

• Related influential characteristics, such as fairness, ethical transparency, safety, and portability were not specified in the proposed model. These attributes may have implications in other domains where ethical and regulatory constraint is high, such as healthcare, education, or public governance.

  Artificial intelligence system reliability and knowledge identity: A model for knowledge workers in knowledge management environments · 2026 · DOI

• Beyond determining how results should be presented, it remains an open question how continuous monitoring of interactions Frontiers in Aerospace Engineering 20 frontiersin. 0—Recent advances, Architecture, and open challenges,” in IEEE consumer electronics magazine 12, 29–37.

  “Interaction Twin in the middle”: a distributed digital twin architecture to model team interactions and dynamics for deep space missions · 2026 · DOI

• Although this paper presents some intriguing findings, there are also some limitations that need to be acknowledged. (1) The limitation in the participant sample may impact the generalizability of the findings across different traffic cultures and demographic groups. Future research should recruit a more diverse participant pool to enhance the robustness of the results, for example, by including both younger and older pedestrians (Dommes, 2019). (2) Due to the limited availability of real AVs, the risks associated with pedestrian-AV interactions, and the practical challenges of manipulating AV platoons in real-world settings, this experiment employed a video-based simulation paradigm. Although meaningful conclusions and insights were drawn, this experimental paradigm may lower pedestrians' risk perception, thus affecting their information perception and decision-making process. Therefore, the results of this study need to be validated in future real-world studies. (3) This study selected a relatively abstract eHMI modality in the form of light. Although abstract eHMI (light-based eHMI that employs traffic metaphors or signals to communicate intuitively through a non-concrete, open-to-interpretation interface (Dey et al., 2020)) has been shown to enable pedestrians to comprehend its meaning, particularly after multiple encounters (de Clercq et al., 2019). However, whether other forms of eHMI, such as text-based, anthropomorphized, or symbolic eHMI, can improve the suboptimal performance of abstract eHMI at partial penetration condition remains to be further investigated. (4) Another limitation is that participants may occasionally miss some deviant stimuli or miscount regular stimuli as deviants in the oddball counting task, which results in an overall count that does not accurately reflect whether the participant correctly captured each deviant stimulus. Future experiments could address this by requiring participants to press a button in response to each deviant stimulus to obtain more precise reaction data. FUNDING STATEMENT This work was supported by the 2025 Key Technological Innovation Program of Ningbo City under Grant No. 2022Z080, and the National Natural Science Foundation of China under Grant No. 72401003. DISCLOSURE STATEMENT The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. DATA AVAILABILITY STATEMENT The data that support the findings of this study are available from the corresponding author upon reasonable request. INFORMED CONSENT STATEMENT All participants have signed informed consent forms permitting the use of their data and images in the publication. The data and images have been anonymized to ensure privacy protection.

  Neuroergonomic evaluation of risk-warning eHMI penetration rates in vehicle platoons: effects on pedestrians’ mental workload, situation awareness, and gap acceptance · 2026 · DOI

• A comprehensive review of influential SA factors, SA components, and SA assessment methods will facilitate driving safety in conditionally AVs but is still lacking.

  Driver Situation Awareness for Regaining Control from Conditionally Automated Vehicles: A Systematic Review of Empirical Studies · 2024 · DOI

• Five ironies of AI are presented including difficulties with understanding AI and forming adaptations, opaqueness in AI limitations and biases that can drive human decision biases, and difficulties in understanding the AI reliability, despite the fact that AI remains insufficiently intelligent for many of its intended applications.

  Ironies of artificial intelligence · 2023 · DOI

• The results of drivers’ cognitive workload were inconclusive, which might be due to the individual and recall biases in subjective measures that could not capture subtle differences in workload during takeover requests.

  Effects of unreliable automation, non-driving related task, and takeover time budget on drivers’ takeover performance and workload · 2022 · DOI

• One question that remains unexplored is: what are the implications of reliability and transparency on trust calibration for human-swarm interaction? We investigate this research question in the context of human-swarm interaction, as swarm systems are becoming more popular for their robustness and versatility.

  The reliability and transparency bases of trust in human-swarm interaction: principles and implications · 2020 · DOI

• BACKGROUND: To increase the use of partially or highly automated driving efficiency and to improve safety, some studies have addressed trust in driving automation and training, but few studies have focused on FAD.

  Fully Automated Driving · 2015 · DOI

• Clinical and organizational researchers alike have studied the effectiveness of self-monitoring as a performance improvement strategy, but further work is needed to determine the power of such interventions for improving safe behavior and to explore the best practices for using such processes with lone workers.

  Behavior-Based Safety and Working Alone · 2001 · DOI

Most-cited papers in Human-Automation Interaction and Safety

• Complacency and Bias in Human Use of Automation: An Attentional Integration · Human Factors The Journal of the Human Factors and Ergonomics Society · 2010 · 1,109 citations
• Automaticity: A Theoretical and Conceptual Analysis. · Psychological Bulletin · 2006 · 901 citations
• What drives the acceptance of autonomous driving? An investigation of acceptance factors from an end-user's perspective · Technological Forecasting and Social Change · 2020 · 302 citations
• Text Messaging During Simulated Driving · Human Factors The Journal of the Human Factors and Ergonomics Society · 2009 · 292 citations
• The future of artificial intelligence at work: A review on effects of decision automation and augmentation on workers targeted by algorithms and third-party observers · Computers in Human Behavior · 2021 · 279 citations
• Supporting Human-AI Teams:Transparency, explainability, and situation awareness · Computers in Human Behavior · 2022 · 249 citations
• Trusting Automation: Designing for Responsivity and Resilience · Human Factors The Journal of the Human Factors and Ergonomics Society · 2021 · 236 citations
• Human confidence in artificial intelligence and in themselves: The evolution and impact of confidence on adoption of AI advice · Computers in Human Behavior · 2021 · 208 citations
• Fully Automated Driving · Human Factors The Journal of the Human Factors and Ergonomics Society · 2015 · 205 citations
• Towards cooperative guidance and control of highly automated vehicles: H-Mode and Conduct-by-Wire · Ergonomics · 2014 · 183 citations

Most recent work

• Antecedents and outcome of situation awareness: a meta-analysis of presence and effect size · Ergonomics · 2026
• Accessibility Geometry Reference Stack: Reference C - Relational and Coupling Variable Register · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Human AI trust modeling in cognitive systems via ensemble learning and advanced feature engineering · Discover Artificial Intelligence · 2026
• AI-augmented decision-making in face matching: comparing concurrent and non-concurrent advice presentation · Cognitive Research Principles and Implications · 2026
• What is Wrong With Automation Bias? · Philosophy & Technology · 2026
• Environmental context is associated with differences in driver behavior around partial automation alerts · Journal of Safety Research · 2026
• Neuroergonomic evaluation of risk-warning eHMI penetration rates in vehicle platoons: effects on pedestrians’ mental workload, situation awareness, and gap acceptance · Scientific Reports · 2026
• Satisfaction in human-DSS interaction is interactively modulated by broad DSS representations and single interactions · Frontiers in Artificial Intelligence · 2026
• Artificial intelligence system reliability and knowledge identity: A model for knowledge workers in knowledge management environments · Knowledge Management & E-Learning An International Journal · 2026
• Of (AI) Machine and Human (Labour): An Integrated Nexus of Work Operating System Architecture for Orchestrating Human–AI Collaborations · Applied Cybersecurity & Internet Governance · 2026

Related topics in Psychology