Vehicles that offer both conditionally automated driving (SAE Level 3, in which the driver may disengage and perform non-driving-related tasks) and partially automated driving (SAE Level 2, in which the driver must continuously monitor the system) create a risk that drivers lose track of the currently active mode. A 2017 simulator study found that, depending on the type of non-driving-related task, shifting between these two automation levels causes a loss of mode awareness and more mode errors than operating a single mode, and that an adapted human–machine interface (HMI) did not restore mode awareness. The subsequent literature leaves the central asks open: no systematic account exists of the specific circumstances (transition direction, task modality, exposure and reliance, simulator versus real road) under which such shifts produce mode errors, no validated HMI concept demonstrably maintains mode awareness across them, and even the measurement of mode awareness lacks a gold standard. The problem is to answer these linked empirical and explanatory questions, including what role HMI design plays in them.

Background

Driving-automation systems are classified by the engineering standard SAE J3016 (published by SAE International, formerly the Society of Automotive Engineers) into levels. Under partial automation (Level 2), the system controls both lateral and longitudinal vehicle guidance, but the driver must permanently monitor traffic and system performance and must intervene even without a warning when the system reaches a limit. Under conditional automation (Level 3), the system can detect its own limits and requests the driver to take over within a time frame, so the driver is allowed to disengage from the driving task and to perform non-driving-related tasks (NDRTs) such as reading or watching videos. Because vehicles increasingly combine systems of both levels, the driver must at every moment know which mode is active and which responsibilities it entails. Mode awareness, a concept originating in Sarter and Woods's analyses of mode error in supervisory control of automated cockpits, denotes the operator's knowledge of the currently active automation mode and of its capabilities and duties; a mode error is behavior that is appropriate for the assumed but not for the actually active mode — for example, engaging in an NDRT while Level 2 is active and monitoring is required. Feldhütter, Segler and Bengler (2017) studied this in a fixed-base driving simulator (a driving simulator with a stationary, non-moving cockpit) with 49 participants who experienced one stage with alternating Level 3 and Level 2 driving and another stage with Level 2 only, crossing this with a cognitive-auditive and a visual-motoric NDRT and with an adapted human–machine interface (HMI) designed to support mode awareness. Shifting between the two modes led, depending on NDRT type, to a loss of mode awareness and to more mode errors than single-mode operation, while the adapted HMI showed no enhancing effect; the automation's low error rate (10%) limited statistical power for undetected system errors, the safe simulator setting may have encouraged risk-taking, and generalization to other automation-level combinations and HMI configurations remained undetermined. A follow-up experiment by the same group (Feldhütter, Härtwig, Kurpiers, Mejia Hernandez and Bengler, 2018) showed that the visual-attention decrement after changing from Level 3 to Level 2 is better explained by overreliance — misplaced trust built through experience with a very reliable system — than by mode-awareness loss per se, and that interposing a short manual-driving phase did not act as the hypothesized countermeasure. Subsequent work clarified measurement but not closure: Kurpiers, Biebl, Mejia Hernandez and Raisch (2020) documented that no gold standard for measuring mode awareness exists and proposed a multi-pillar protocol combining gaze-based attention ratios, mental-model interviews, and NDRT engagement; Forster, Geisel, Hergeth, Naujoks and Keinath (2020) showed NDRT engagement itself can serve as an online, non-intrusive indicator, rising with exposure and more strongly under Level 3 than Level 2; and Plum and Shi (2025) still report the robustness of mode-awareness assessment as unsettled. On the design side, alternatives to the failed adapted HMI have been explored — gaze-direction attention requests (Kurpiers, Lechner and Raisch, 2019), function-centred versus level-centred control interfaces (Eom and Lee, 2022), visual cues in the steering wheel (Diederichs et al., 2022), and ambient light displays spanning manual, Level 2 and Level 3 driving (Mirnig et al., 2023) — with partial successes (e.g., ambient lights on the steering wheel and below the windshield were rated useful while footwell lights were not; function-centred interfaces halved confusion rates) but also recurring awareness–behavior gaps in which participants knowingly continued NDRTs under Level 2 monitoring requirements. A review by Dönmez Özkan et al. (2021) concluded that existing mode-awareness interfaces are simple or outdated and difficult to extend without overloading the driver. Field evidence has begun to appear: Britten and Perez (2022) measured mode awareness after on-road transitions in a Wizard-of-Oz vehicle (a real car in which automation is secretly operated by an experimenter, used to study realistic behavior safely), and Kim, Novakazi and Karlsson (2025) observed genuine mode-transition and mode-awareness errors — unintended Level 2 deactivations and confusion after Level 3 disengagement — in an on-road vehicle offering manual, Level 2 and Level 3 modes; Kim et al. (2024) additionally documented, in an 838-respondent survey, systematic misalignment between drivers' expected and actual control responsibilities after interventions under both levels. These strands jointly define the state of the art against which the unresolved question below must be judged.

Problem Statement

Determine, based on the existing empirical literature, under which circumstances shifting between conditionally automated driving (SAE Level 3) and partially automated driving (SAE Level 2) within one vehicle degrades drivers' mode awareness and produces mode errors, and for each circumstance state the studies that support it and the limits on its generalisation. The circumstances to determine are five, each independently answerable and jointly posed:

  1. the direction of the transition;
  2. the modality of the non-driving-related task (auditive-cognitive versus visual-motoric);
  3. the roles of exposure duration, automation reliability and resulting overreliance versus genuine mode confusion;
  4. whether and how human–machine interface (HMI) design (e.g., conspicuous distributed feedback such as ambient light, attention requests, adaptive interruption of non-driving-related tasks, or function-centred control interfaces) affects mode awareness across such shifts relative to a baseline interface;
  5. whether findings transfer from driving simulators to real-road traffic.

The absence of a gold-standard measure of mode awareness is a limitation inherited by every part, not a separate mandated deliverable; each part's measurement basis and the limits it imposes are to be stated, and establishing or validating a new measurement standard is out of scope. The common objective of these linked questions is that a vehicle offering both automation levels can be operated safely: the circumstances account explains when and why supervision fails at level shifts, and the HMI findings show what an interface can contribute to preventing it, each informing and calibrating the other. Whether comparable effects arise for other automation-level combinations (e.g., Level 2 and Level 4) and other HMI configurations remains open in the subsequent literature.

References

  1. Feldhütter, A., Segler, C., & Bengler, K. (2017). Does Shifting Between Conditionally and Partially Automated Driving Lead to a Loss of Mode Awareness? Advances in Human Aspects of Transportation, pp. 730–741. DOI: 10.1007/978-3-319-60441-1_70. https://doi.org/10.1007/978-3-319-60441-1_70
  2. Feldhütter, A., Härtwig, N., Kurpiers, C., Mejia Hernandez, J., & Bengler, K. (2018). Effect on Mode Awareness When Changing from Conditionally to Partially Automated Driving. Proceedings of the 20th Congress of the International Ergonomics Association, pp. 314–324. DOI: 10.1007/978-3-319-96074-6_34. https://doi.org/10.1007/978-3-319-96074-6_34
  3. Kurpiers, C., Biebl, B., Mejia Hernandez, J., & Raisch, F. (2020). Mode Awareness and Automated Driving—What Is It and How Can It Be Measured? Information, 11(5), 277. DOI: 10.3390/info11050277. https://doi.org/10.3390/info11050277
  4. Kurpiers, C., Lechner, D., & Raisch, F. (2019). The influence of a gaze direction based attention request to maintain mode awareness. Proceedings of the 26th International Technical Conference on the Enhanced Safety of Vehicles (ESV), Eindhoven, The Netherlands, June 2019.
  5. Forster, Y., Geisel, V., Hergeth, S., Naujoks, F., & Keinath, A. (2020). Engagement in Non-Driving Related Tasks as a Non-Intrusive Measure for Mode Awareness: A Simulator Study. Information, 11(5), 239. DOI: 10.3390/info11050239. https://doi.org/10.3390/info11050239
  6. Dönmez Özkan, Y., Mirnig, A. G., Meschtscherjakov, A., Demir, C., & Tscheligi, M. (2021). Mode Awareness Interfaces in Automated Vehicles, Robotics, and Aviation: A Literature Review. Proceedings of the 13th International Conference on Automotive User Interfaces and Interactive Vehicular Applications (AutomotiveUI '21), pp. 147–158. DOI: 10.1145/3409118.3475125. https://doi.org/10.1145/3409118.3475125
  7. Mirnig, A. G., Gärtner, M., Wallner, V., Demir, C., Dönmez Özkan, Y., Sypniewski, J., & Meschtscherjakov, A. (2023). Enlightening mode awareness. Personal and Ubiquitous Computing, 27(6), 2307–2320. DOI: 10.1007/s00779-023-01781-6. https://doi.org/10.1007/s00779-023-01781-6
  8. Britten, N., & Perez, M. A. (2022). On-Road Assessment of Driver Mode Awareness of Assisted and Automated Driving. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 66(1), 357–361. DOI: 10.1177/1071181322661145. https://doi.org/10.1177/1071181322661145
  9. Kim, S., Novakazi, F., van Grondelle, E., van Egmond, R., & Happee, R. (2024). Who is performing the driving tasks after interventions? Investigating drivers' understanding of mode transition logic in automated vehicles. Applied Ergonomics, 121, 104369. DOI: 10.1016/j.apergo.2024.104369. https://doi.org/10.1016/j.apergo.2024.104369
  10. Kim, S., Novakazi, F., & Karlsson, I. C. M. (2025). Is conditionally automated driving a bad idea? Observations from an on-road study in automated vehicles with multiple levels of driving automation. Applied Ergonomics, 129, 104617. DOI: 10.1016/j.apergo.2025.104617. https://doi.org/10.1016/j.apergo.2025.104617
  11. Eom, H., & Lee, S. H. (2022). Mode confusion of human–machine interfaces for automated vehicles. Journal of Computational Design and Engineering, 9(5), 1995–2009. DOI: 10.1093/jcde/qwac088. https://doi.org/10.1093/jcde/qwac088
  12. Diederichs, F., Muthumani, A., Feierle, A., Galle, M., Mathis, L.-A., Bopp-Bertenbreiter, V., Widlroither, H., & Bengler, K. (2022). Improving Driver Performance and Experience in Assisted and Automated Driving With Visual Cues in the Steering Wheel. IEEE Transactions on Intelligent Transportation Systems, 23, 4843–4852. DOI: 10.1109/TITS.2022.3162522. https://doi.org/10.1109/TITS.2022.3162522
  13. Plum, L., & Shi, E. (2025). Raising Awareness for Confusion - Stimulating the Discussion About Robustness of Mode Awareness Assessment in Automated Driving. AHFE International. DOI: 10.54941/ahfe1007023. https://doi.org/10.54941/ahfe1007023
  14. Sarter, N. B., & Woods, D. D. (1995). How in the world did we ever get into that mode? Mode error and awareness in supervisory control. Human Factors, 37(1), 5–19. DOI: 10.1518/001872095779049516. https://doi.org/10.1518/001872095779049516