CONNECTED & AUTONOMOUS VEHICLES
RISK VERSUS BENEFIT
New vehicle regulations for SAE level 3
Automated Vehicles are widely considered to have huge economic and safety benefits and around the world industry is racing to be first to market. Meanwhile, regulators are working hard to try to accommodate the technology safely, but the implications of changes can be significant. Careful analyses are required, as Iain Knight, Prof Alan Stevens, Helen Viner and Martin Lamb explain
The Society of Automobile Engineers (SAE, 2018) have defined levels of automation from 0 to 5, where 0 is no automation and 5 is a vehicle that can be automatically driven on-road in any area a human driver can. Level 1 and 2 represent driver assistance systems; they automate certain elements of the mechanical driving task but the human driver remains responsible for constant monitoring of the road, traffic and driving system behaviour. Level 4 and 5 clearly represent automated driving of sufficient technical standard that, at least in certain defined driving circumstances (known as operational design domains), the driver is not required to have any input to the driving task. At this level, the “driver” can safely engage in other tasks while leaving the system to operate the vehicle.
In the middle-ground, driving automation at SAE level 3 has proved controversial. At level 3, the vehicle will simultaneously control acceleration, braking, and steering. It will also be responsible for monitoring the road around the vehicle and executing all the detailed path planning required. This means that the driver no longer needs to be responsible for the constant monitoring of the driving environment or the supervision of the driving system operating the vehicle. Effectively, the design intent of an SAE level 3 system is that the driver will be free to do other things while it operates the vehicle. However, this type of system may not have sufficient built in redundancy to always find and reach safe harbour by itself if a part of the system fails.
Similarly, it might not be capable of dealing with all driving circumstances it might encounter on a given section of road. That is, its operational design domain may be limited to very specific circumstances and when the vehicle is no longer in those circumstances, it will expect the driver to take over, potentially at quite short notice. The SAE (SAE, 2018) suggests at level 3 the automated driving system will be able to continue to operate the vehicle ‘for at least several seconds’ after an intervention request.
“Its operational design domain may be limited to very specific circumstances and when the vehicle is no longer in those circumstances, it will expect the driver to take over, potentially at quite short notice”
The challenge comes from the question of whether a driver that has not needed to pay attention to the driving will be ready to take over at short notice. Research has suggested that drivers will want to undertake a range of activities including, for example, engaging with other vehicle occupants, looking at scenery, viewing and listening to video and/or audio material, reading, personal grooming and sleeping. This last activity is particularly interesting; it has been suggested not only by older drivers but, with a 24/7 culture becoming more common, by a range of drivers of all ages. The key point about these activities is that the driver may be partially or fully disengaged with what is going on around the vehicle i.e. “out of the loop” in terms of vehicle control.
A driver of an SAE L3 system may, therefore, have to go from a potentially disengaged state to a fully in-control state in a very short timeframe and there are two key factors here: the time taken and quality of the take-over. The time taken depends not only on the driver’s state at the time and personal characteristics (such as reaction time) but also on how the take-over request is presented to the driver.
Research has suggested that takeover can be enhanced if the request is presented through multiple modes (visual, auditory and haptic), if the request is familiar and easily understood, and if it can direct the driver’s attention to where it is most needed. Some researchers have distinguished different features of the take-over timeline from initial reaction to the request, first glance at the external scene, resuming the driving position (hands at wheel, feet at pedals), first control action (steering or pedal movement) etc. The quality of take-over is more difficult to measure and a range of techniques has been developed for use in simulators and field trials. Essentially the driver needs to recover good “situational awareness”; that is the driver needs to perceive and then understand the external situation and be able to predict how it is likely to evolve over the coming seconds.
Importantly, there is no consensus in the human factors community on how much time is needed for the driver to take over vehicle control following disengagement with the driving task (the literature reports findings from around 2 seconds to over 30). In part, this is because of the wide variety of use cases reported in simulations and track trials, different measurement techniques and different interpretations of findings. It also clearly depends on the driver, their degree of alertness at the time and what they are doing when take-over is required.
“The driver needs to recover good “situational awareness”; that is the driver needs to perceive and then understand the external situation and be able to predict how it is likely to evolve over the coming seconds”
So, some difficult challenges lie ahead. But that’s ok, right? Because there is plenty of time to sort it all out, isn’t there?
Well, way back in 2017, Audi announced (Audi, 2017) the arrival of ‘the world’s first system that enables highly automated driving at level 3’. The system is named ‘Traffic Jam Pilot’. On motorways, the system can take over the driving in traffic jams and slow-moving traffic at up to 60 km/h (37 mph). Audi stated that ‘with the traffic jam pilot, the driver can relax. In this specific traffic situation, they can take their hands off the steering wheel permanently and, depending on national laws, focus on a different activity that is supported by the vehicle infotainment system’. The system also monitors the driver to assess whether he or she is tired or asleep and, if that is found to be the case, a multi-stage warning will be issued. When the speed rises above 60 km/h and/or the traffic around the vehicle breaks up, then the system will immediately ask the driver to take charge again. If the driver fails to respond to the request and subsequent warnings, the vehicle will brake to a standstill.
It is clear that Audi have attempted to address concerns around the driver role and has put technical solutions in place to mitigate the risk. However, many questions remain, for example:
Audi suggest that the system will allow the driver to engage with the vehicle infotainment system, which implies that they will not be completely free to let the vehicle take the strain. Some evidence shows that drivers actively engaged with vehicle systems can respond more quickly than those doing nothing but there is little independent assessment of how it compares to engagement with nomadic devices (phones, tablets, books etc.), which would presumably be one attractive choice to drivers. Whether the general driving public or even the enforcement authorities will understand this distinction is an open question.
Basic forms of monitoring steering and pedal inputs for patterns characteristic of fatigue and distraction have been in some production vehicles for years. However, there is relatively little independent evidence of how effective the proposed system is, how well it monitors for distraction other than fatigue, whether there is any system to prevent secondary tasks not involving the vehicle infotainment system or the extent to which it might suffer false detections of tiredness.
Is an SAE definition of the situation the vehicle should achieve to minimise risk in the event the automated system cannot continue. In the case proposed by Audi, this appears to be a simple stop in lane. If so, can this really be considered minimum risk, when other traffic is accelerating? The risk of collision from behind is clear, the UK Highway Code says that vehicles should only stop in lane in an emergency and the actions of drivers trapped behind the stopped vehicle could also be risky.
On the one hand, human driven cars do stop in lane as a consequence of vehicle failures and collisions are rare. However, the effect on frequency of events is key. Catastrophic failures of current vehicles are rare and most drivers have never been stranded in a live lane. However, traffic jam pilot systems could exit their operational design domains multiple times on every congested motorway trip and every exit is a potential ‘stop in lane’ if the driver is not sufficiently alert.
The technology for traffic jam pilot was ready some time ago; it is only legislation that is stopping its use in production vehicles. While Audi may be first, it is certain others will be working on similar systems. In the UK, vehicles must be ‘approved’ before they can be registered for use on the road and much of the relevant technical regulation is globally harmonised through the United Nations Economic Commission for Europe (UN ECE) and a task force is currently busy considering all of these complex issues and more.
Developing a regulation to permit this type of traffic jam automation, generically referred to as a low speed automated lane keeping, is seen as the next steps after recent amendments to steering regulations permitted automated control of steering activation in a wider range of circumstances. Work is expected to be completed next year. The current (early) draft appears likely to permit the Audi system with little, if any, functional modification.
The complete regulation will allow such systems to be sold in the UK but will not give drivers permission to take their hands off the wheel. The use of vehicles is a national matter and would require changes to UK regulations and the Highway Code. Once amendments are permitted to allow drivers to do secondary tasks, it is then likely they would have to be designated as ‘automated’ under the recent Automated and Electric Vehicles Act, which governs the change to insurance arrangements for automated vehicles. The insurance industry would be sure to have an influential view.
“In the UK, vehicles must be ‘approved’ before they can be registered for use on the road and much of the relevant technical regulation is globally harmonised through the United Nations Economic Commission for Europe”
In the meantime, the members of ITEN will continue work to objectively inform all of the stakeholders in this debate of the likely implications of technologies, given a range of different options for regulatory constraints.
We are keen to hear the views of the Thinking Highways community. What constraints do you think are the minimum acceptable for this sort of system? What data could help to better quantify the risks? What research can we do to provide better answers to the key questions raised? Automation is certainly coming, but a lot of work is still required to maximise the benefits and avoid the risks.
Audi, 2017. Audi A8 - Audi AI traffic jam pilot. s.l.:Audi AG Press release.
SAE, 2018. Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles, s.l.: Surface Vehicle Recommended Practice J3016, SAE International .
The International Transport Experts Network (ITEN)
ITEN is a network of small, independent transport consultancies providing evidence-based research, consultancy and strategic advice to public and private clients. Working together we provide multi-disciplinary expertise, including:
- Vehicle safety
- Highway infrastructure and operations
- Railway systems
- Future transport systems and technology strategy
- Reducing harmful impacts of transport
Iain Knight is director and principle engineer at Apollo Vehicle Safety.
Prof. Alan Stevens is an independent ITEN Associate
Helen Viner is director of Enodamus
Martin Lamb is the founding director of Maple Consulting