Recent research in the UK, sponsored by the Institute of Advanced Motorists, has focused on developing and evaluating a driver training program for Level 4 automated vehicles (Merriman, 2021a, Merriman et al., 2021b). During the Covid-19 pandemic, 96 drivers undertook an online video-based training program based on current UK Hazard Perception training materials, with the specific aim of improving drivers mental models for understanding when level 4 automated can be activated. It was found that the online training programme, in combination with an owner’s manual, led to a greater improvement in drivers mental models for when automation can be activated compared to the owner’s manual in isolation (which is the current, non-mandated, approach to training when someone purchases a vehicle with capacity for automation). Interestingly, it was found that drivers’ trust in automation reduced after undergoing training. This suggests that drivers initially had misplaced high trust in the capabilities of automated vehicles, which, without training, has the potential to influence their unsafe interactions with them on the road.
An extension of the online training program has been evaluated in a driving simulator at the University of Southampton with 45 drivers. It was found that drivers who read the owner’s manual (akin to current training when purchasing a vehicle with automated features) or underwent the Level 4 driver training package made more correct decisions about whether it was safe or not to activate the automation on the simulated tasks compared to drivers who received no training. Drivers who only read the owner’s manual experienced a greater mental demand when making their decisions compared to drivers who underwent the Level 4 driver training package. An analysis on the drivers’ reasoning for their decisions revealed that drivers who underwent the Level 4 driver training package had more appropriate and comprehensive mental models for when the automation can be activated compared to drivers in the other two training conditions. It is recommended that a comprehensive and standardised driver-automation training programme is developed and rolled out. This should be mandated for owners of vehicles with automation capability, in the same way current driver training is.
Research generally focuses on how driver behaviour may change as a result of increasing the level of automation, however a much broader insight into the entirety of automated driving system to reveal more fundamental systemic sociotechnical concerns is required. Banks et al. (2019) conducted a system-based analysis, which highlighted a lack of top down influence within the system network which means that lower tiers lack appropriate support and guidance. Vehicle manufacturers, to some extent, have been left to their own devices when it comes to designing, testing and marketing their systems. This is a problematic approach but the UK is not the only country facing these issues. The United States for instance has been testing automated vehicles for some years. This has largely been driven by the companies located within Silicon Valley (i.e., at an ‘industrialist’ level) rather than through the adoption of a top-down approach (i.e., Government led). Whilst this has enabled developments within the field to accelerate at a rapid pace, we are also now beginning to see the negative aspects of automated design (e.g., driver complacency, automation misuse and ethical dilemmas).For example, in a recent analysis on five Level 2 collisions that occurred in the USA between 2016-2019, which found that a consistent pattern of driver-behaviour emerged across all five collisions (Merriman et al., 2021b). In general, the drivers had an incorrect mental model (e.g., knowledge and understanding) for the automation and over-trusted the automation. In combination with the ‘task underload’ caused by the automated driving environment, in which the drivers were not actively involved in the driving task, they did not continually attend to the road environment or the operation of the vehicle. Instead some drivers performed a secondary task (e.g. watching a film/television programme, eating or playing a game). This resulted in drivers believeing that the automation was more reliable than it actually was and over-relied on it. This research highlighted the limitations with current driver training for automated vehciles (or lack of) and defined the competencies that need to be targeted in future trianing programmes, namely workload, speed of prcoessing, mental models, trust, attention and memory, situation awareness, procedureal skills, hazards and risk perception, and siitutudes and personality).
It is clear that greater ‘top-down’ (i.e. International and National Committees and Governments) influence is required to evoke the change necessary to ensure that automated systems are properly tested and verified. Banks et al., (2019) propose that a combined top-down, and bottom-up, sociotechnical systems approach is adopted. Such an approach recognises the significant progress made by the intermediate tiers of the system (e.g., from ‘Industrialists’) in producing the evidence required to update, amend and event create new design standards and protocols to enable certification will be essential. This evidence needs to be acted on by International and National Committees to generate standards and polices that will go on to inform new government policies and legislation across countries. This in-turn will inform regulators (i.e. top down) and so on. Suggestions for targeted recommendations at each level of the automated driving system are provided below
Table 1. Targeted recommendations for each level of the automated driving system (taken from Banks et al., 2019)
1. Provide clearly worded legislation surrounding the design, use, testing of and implementation routes of automation to vehicle manufacturers
a. Clearly outline the role of the driver and their responsibilities
3. Provide clear guidance to industrialists and resource providers and ensure appropriate enforcement
5. Provide evidence of thorough testing highlighting the potential risks involved with using the system. This will facilitate the exploration of systemic actions that could minimise risk
8. Consider additional training programmes for drivers of automated vehicles
10. Drivers must understand the limitations and capabilities of the automated system in use. They must have a clear understanding of their roles and responsibilities
Equipment and Environment
11. Activities conflicting with the driving/new monitoring task must be understand by other element of the system (e.g. Police) who must be able to enforce laws relating to new devices in the vehicle
Merriman et al. (2021a; 2021b) has demonstrated that without training, drivers do not have an appropriate mental model (knowledge and understanding) for when it is safe to activate a Level 4 automation. If drivers do not receive training, they are more likely to activate the automation in unsafe road conditions.
Merriman, S.E. 2021a. Do you know what an AV thinks is a hazard? The Development and Evaluation of an Online Video-Based Training Programme for Drivers of Level 4 Avs. A seminar given to the Transportation Rearch Group, University of Southampton, 10/12/2021
Merriman, S. E., Plant, K. L., Revell, K. M., & Stanton, N. A. (2021b). What can we learn from Automated Vehicle collisions? A deductive thematic analysis of five Automated Vehicle collisions. Safety Science, 141, 105320. doi:https://doi.org/10.1016/j.ssci.2021.105320