Transition of control in automated vehicles

Overview of solutions and acceptance by industry
Auteur(s)
Jansen, R.J.; De Goede, M.; Van Grondelle, E.D.
Jaar

Until recently the driver of a vehicle was the only agent capable of (safe) execution of the driving task. With the arrival of Automated Driving Systems (ADSs), the execution of (a part of) the driving task has been transferred from a human agent to a non-human agent: ‘the vehicle’. At the same time, the human agent is still obliged take over control in case the ADS reaches the limits of its Operational Design Domain (ODD). From a human factors (HF) perspective, this transition of control (TOC) of (part of) the driving task is a critical moment. In literature, a great deal of research on this topic and on solutions for the TOC has been published.

The Dutch Ministry of Infrastructure and Water Management (IenW) is exploring how human factors requirements originating from scientific literature can be added to the type approval process of vehicles (e.g., passenger cars, vans, buses, trucks) equipped with ADS. According to IenW, interaction between ADS and the driver plays an important role in relation to safe use of these systems. Therefore, evaluation of the interaction between ADS and the driver may become part of the type approval process. In a previous project commissioned by IenW, SWOV found that the automative industry is critical towards user-centred design (UCD) process audits, in which human factors requirements play a key role (e.g., representatives would like to see proof that applying UCD yields better products, that workload does not increase, and that prescribed processes are beneficial to their own business processes).[1] A critical perspective may also be present towards the potential outcomes of such a process (e.g., TOC solutions).

IenW has asked SWOV to investigate whether TOC solutions reported on in scientific literature would be solutions that the industry chooses or is willing to choose to incorporate in their vehicle systems. Therefore, the aim of the current project is to see how ideas on and developments in TOC, including TOC solutions, in scientific literature relate to relevant thoughts and solutions in the industry.

The research questions answered within this project are:

  1. Which scientifically substantiated effective TOC solutions can be found in scientific literature?
  2. What does the industry (e.g., car manufacturers) think about effective TOC solutions, as described in scientific literature?
  3. What human factors issues in relation to TOC remain unsolved, despite alleged solutions?

The approach aimed at answering the previously defined research questions consists of three phases:

  1. Literature review. A screening of 40 scientific papers and deliverables of 10 EU projects on TOC was carried out to end up with a maximum of 30 documents that describe a TOC solution.
  2. Selection of solutions. The project team formulated scientific quality requirements (e.g., replicability, reliability, internal validity, and external validity) and experiment outcome requirements (e.g., a positive effect on safety performance indicators and/or user experience measures), and selected a set of 5 solutions on that basis.
  3. Evaluation by industry. The subject of TOC as well as the selected solutions were discussed in interviews with nine employees at four different Original Equipment Manufacturers (OEMs).

Literature review

The literature review of TOC solutions consisted of the screening of 40 selected scientific papers, as well as 10 EU project deliverables. For further reading, 19 documents were selected. The resulting TOC solutions were evaluated on scientific quality as well as their experimental outcomes. Some studies showed limitations in terms of one or more of the following observations:

  • External validity was assessed on, amongst others, the similarity between results obtained through a driving simulator setup and results obtained through on-road testing. Several driving simulator studies used outcome measures of which a meta study suggests that the measured values (e.g., lateral position, steering, braking in case of driving simulator setups with low fidelity) differ from when the same measures would have been obtained in real-life driving.
  • Of several outcome measures used in driving simulator studies the external validity (e.g., as described above) is currently unknown. This does not mean that the obtained results are invalid, but at the moment there is no way of knowing how the results will generalize to real-life driving.
  • TOC solutions are typically investigated in isolation, whereas in real-life driving drivers also interact (sometimes simultaneously) with other car-systems, such as navigation, entertainment systems and ADAS.
  • Response time (operationalised in various ways, e.g., time from a takeover request (TOR) onset to hands on steering wheel, time from TOR onset to brake pedal use) is most frequently used to make a statement on the success of a required transition from automated driving system to driver, despite indications that driver reactions are of poorer quality when they are faster, due to time required to build up situational awareness (Gold et al., 2013).[2]

Nonetheless, several studies were assessed to be of good scientific quality and the corresponding TOC solutions showed an improvement in terms of safety performance indicators and/or user experience measures. Based on these scores as well as an expert opinion on practical feasibility of these solutions for industry, a subset of five solutions was selected to present to and discuss with employees within the automotive industry, namely:

  1. Time budget. This feature indicates the amount of time that is left within the current automation level and/or the time that is left between a takeover request and the actual takeover.
  2. MEDIATOR takeover ritual: a fixed ritual consisting of ambient lighting, a time budget indicator and auditory and visual signals to indicate an upcoming takeover, with increasing levels of urgency in case the driver does not undertake any action.
  3. Ambient lighting, which means communicating current and/or upcoming automation levels with different colours of ambient lighting.
  4. Effective user interface features in a transition of control, that is: auditory signals, urgency and context information.
  5. Assisted and Partial Takeover: the combination of partial takeovers that allow the driver to take over either the lateral or longitudinal control and assisted takeovers in which the driver is monitored after takeover and is assisted automatically through soft braking and additional warnings and emergency braking if the time-to-collision falls below a critical level.

Interviews

Four interviews were conducted with nine persons, working at four different OEMs. The interviewees were asked about: ADSs and TOC in general, their opinions on the five selected solutions, their ideas on human factors issues related to TOC, standardisation, differences between science and industry as well as the external validity of driving simulator studies.

Ideas on ADSs and TOC by industry

The degree to which the interviewees’ companies were involved in the TOC concept differed. Two out of four OEMs have systems in operation which actually transfer responsibility for execution of the driving task from the driver to the car and vice versa (level 3 systems). The differences between the companies did however not entail differences in the way TOC was understood. However, the two OEMs that have AD (automated driving) systems (almost) in operation had more distinct opinions on part of the presented solutions, compared to the other OEMs. One OEM expressed a strong opinion on underlying HF principles, presenting their view on ADS from a user perspective, not from what is technically feasible. In contrast, other OEMs seemed to reason more the other way around: what is technologically feasible and how should we communicate this to the users?

Industry perspective on the selected solutions

The solution concepts that were presented to interviewees were all more or less known to them. Participants did not discuss the complete solution in detail, but picked out one or more aspects which they had specific thoughts about or experience with. The takeover ritual, ambient lighting and general principles of modalities yielded positive reactions from most OEMs. However, for every solution they also expressed areas of concern regarding their implementation, user perception, order and timing of modalities, as well as the traffic context in which it is to be used. The discussion on ADS, TOC and TOC solutions revealed that the defined issues and concerns related to TOC are substantially the same in science and industry. However, hands-on testing in industry brings to light more practical issues and contradictions between on-road tests and tests in a well-controlled experimental setting.

Remaining Human Factors issues

Overall, the participating OEMs indicated that design and implementation challenges do not relate to a lack of Human Factors knowledge. In fact, three OEMs stated that the problem is not the theoretical knowledge about human factors issues (e.g., driver distraction as a construct) , but the way human factors issues are mitigated through design. However, the translation of HF knowledge into system features can be considered as a further refinement of more general underlying principles (e.g: information should not be distracting, information should focus on sufficient situational awareness on the part of the driver when taking over control). Therefore, it can be concluded that the operationalisation of Human Factors principles/guidelines, that is the application of these general guidelines in specific system design, seems to need additional as well as ecologically valid research.

Standardisation

In general, the OEMs stated that the industry is not in favour of standardisation, which they consider to limit innovation and branding opportunities. But interviewees also stated that, from a personal point of view and/or the point of view of a human factors expert, they are in favour of standardisation. It was also indicated that a common ground is especially relevant in the context of different people sharing the same vehicle, as well as one person switching between vehicles .

External validity

Closely related to the issue of the operationalisation of human factor knowledge are OEM concerns of insufficient external validity of driving simulator studies. Not only science, but also OEMs often struggle with the fact that driving simulator studies do not always have sufficient ecological validity in order to translate results to the actual road. Based on the literature review, it can be concluded that many findings in science do not feature much or any (known) external validity. Barriers are legislations as well as ethical considerations to prevent too much risk for participants in on-road testing. The latter barrier implies that there are sometimes no practical alternatives to using driving simulator studies.

Conclusion

This project investigated which scientifically substantiated effective TOC solutions are described in literature, and explored how the automotive industry views these solutions, as well as what human factors issues remain because existing solutions fall short. Overall, knowledge in science and industry on human factors related to TOC do not appear to be far apart. Several TOC solutions in literature were rated as valuable based on scientific criteria and found effects in an experimental setting. Some potentially valuable TOC solutions were discarded due to limitations in the (reported) scientific approach. The possibility that such TOC solutions may in fact have merit indicates that more awareness on (reporting) scientific rigour is needed in the scientific community. Three TOC solutions (e.g., takeover ritual, ambient lighting, general principles about modalities) yielded positive reactions from most OEMs. Yet, for all of these solutions interviewees indicated ‘flaws’ or at least critical considerations, based on 'practical’ experience and knowledge from actual interaction with systems on the road. The interviews also revealed a lack of knowledge from an industry perspective about translating (well-known) human factors principles into applicable design guidelines. The interviewed OEMs seem to consider this knowledge gap as specifically related to each individual system or vehicle. However, it might relate, on a more general level, to the challenge that science faces, that is making human factors concepts measurable and define thresholds to quantify the quality of a solution.


[1].   De Goede, M., Jansen, R.J. & Van Grondelle, E.D. (2022). User-centred design for type approval of AD(A)S; Roadmap towards a process audit. R‑2022‑16, SWOV, The Hague.

[2].   Gold, C., et al. (2013). “Take over!” How long does it take to get the driver back into the loop? In: Proceedings of the Human Factors Ergonomic Society Annual Meeting. p. 1938-1942.

Rapportnummer
R-2024-7
Pagina's
119
Gepubliceerd door
SWOV, The Hague

SWOV-publicatie

Dit is een publicatie van SWOV, of waar SWOV een bijdrage aan heeft geleverd.