NTC submission – autonomous vehicle trials
Monday, January 23, 2017 by Scott Coles
Australia’s National Transport Commission (NTC) has requested submissions for their paper on “National guidelines for automated vehicle trials”. The purpose is to have state and territory road transport agencies working under a common national approach on how they will regulate and support automated vehicle trials to ensure public safety. The NTC wants to provide certainty and clarity, help with management of trials and establish minimum safety standards to raise awareness and acceptance by the public and to protect those using public roads.
From Optalert’s perspective, we see the major problem as safety. When drivers are on public roads trialling partially and conditionally automated vehicles, they face a greater danger of a sleep-related accident, as these vehicles require the driver to be alert, but take away the mental stimulation provided by the act of driving. This could be a formula for disaster.
If accidents are to occur during the trial phase, lives may be lost and the public may irreparably lose confidence in the technology.
Alertness required for autonomous vehicles
Optalert has determined the alertness levels required for each of the SAE automated vehicle classifications.
Figure 1. (Click to view full size)
Autonomous (autopilot) mode in an autonomous vehicle
Autopilot has one key function: it removes the need for a driver to be actively engaged while a vehicle is moving. However, according to the levels of autonomous driving listed in Figure 1, classifications named as level one, two or three still require a driver to be alert and aware of their surroundings. In an urban environment when drivers are in direct control of a vehicle, they are constantly checking for traffic signals, predicting what other cars are going to do, and shifting their vision between the road, other cars, pedestrians, mirrors, and the dashboard.
All these actions increase the cognitive load and can help keep the driver alert, but autopilot removes the need to do these things, and that can lead to cognitive underload. The natural drowsiness state, based on how much sleep the driver has had, is now in danger of dominating.
In an instance where autopilot needs to hand control back to the driver, it’s likely that something has occurred that the autopilot cannot handle. It might be the software has lost track of lane markings or there is an upcoming obstacle that it cannot identify or safely avoid. At this exact moment, the driver needs to be alert and ready to take back control. But, if the driver has been subjected to a prolonged period of mental underload, they may experience signs of drowsiness, which means reaction times are impaired during a period where they need to be immediate.
If there was a way for an autonomous vehicle to detect if a driver was alert and capable of taking control of the car, then different actions could be taken. For example, if the driver was asleep, the car could execute a controlled stop and alert the driver to wake up before an emergency occurs. If the vehicle could detect that a driver is showing the early signs of drowsiness, it could become more responsive and take earlier action in instances where control needs to be returned to the driver. However, until those functions are offered within vehicles, we recommend drowsiness-detection to be mandatory in autonomous vehicle trials.
Minimum safety standards
We are strong advocates of real-time monitoring of alertness/drowsiness in drivers, to prevent them from ‘falling asleep at the wheel’. In the case of autonomous vehicle trials, it is essential systems are in place to warn of the early stages of drowsiness where performance becomes impaired. This should be mandatory for minimum standards of safety.
In all the semi-autonomous modes (up to Level 4), the driver will still need to be able to take control of the vehicle at any time. Even though the Level 3 and Level 4 definitions suggest the driver does not need to pay attention, or is not even required, this is only during defined environments, and this could change, potentially rapidly, at any point in the journey. Although the driver’s level of engagement with actually driving and controlling the vehicle diminishes as the vehicle becomes more autonomous, the requirement to pay attention to the road is still high up until Level 5.
Of all referenced transport agencies, Australia has the largest national distances and needs to be at the forefront of protecting the lives of the test drivers and those around them. We need to be world leaders in setting the level of safety around drowsiness.
There has not been a consistent approach adopted in relation to fitness for duty in existing codes and guidelines. The New Zealand guidelines state the driver must be “unimpaired” (NZ 26 Transport Agency, 2016), while the UK code states that the trialling organisation “should develop robust procedures to ensure that test drivers and operators are sufficiently alert to perform their role and do not suffer fatigue” (UK Department of Transport, 2015). The UK code suggests measures to ensure the driver’s alertness could include setting limits around the total amount of time test drivers or operators perform such a role per day, and the maximum duration of any one test period (UK Department of Transport, 2015). Likewise, the SAE standard for testing automated vehicles provides a standard in regard to the maximum number of consecutive hours a test driver may operate a trial vehicle, which should be included as part of the driver’s or operator’s training (SAE International, 2016).
Determining alertness levels in real time
Drowsiness tests based on questionnaire and usage times will NOT be sufficient, as questionnaires are subjective. Like Blood Alcohol Concentration (BAC), drowsy drivers do not know when they are impaired without technology and drivers may start driving already drowsy due to environmental, physical and or medical factors. Hour-based methods will therefore not be sufficient. In addition, the effect of not having the mental stimulation of driving, but still being required to be alert, has not been quantified scientifically or empirically, i.e. the current guidelines for maximum duration of any rest period is not known.
Fortunately Australia is leading the world in drowsiness detection. Optalert's technology monitors the alertness/drowsiness of individual drivers in real time. Like Blood Alcohol Concentration (BAC), drowsiness can be measured. The Johns Drowsiness Scale (JDS) is one such measure that has been independently validated. This scale ranges from 0 (very alert) to 10 (very drowsy) (Fig 2).
The additional benefit of using drowsiness detection devices, can be the ability to collect data determining the times and locations where drivers participating in the autonomous vehicle
trials have been drowsy. This could be plotted against vehicle control to determine the most at risk locations, drivers and situation of vehicle control. Below is an example of data plotted for average JDS scores and times of the day when they occurred.
Figure 3. (Click to view full size)
Figure 4 is a map which shows the locations where Optalert JDS warnings have been issued to drivers on a particular route. This kind of map can be plotted for any road where JDS scores have been collected.
Optalert’s recommended safety options
As experts in drowsiness detection, we recommend more stringent options for each relevant option within the Safety Management Plan.
4.3 Option2 - Guidelines allow testing without a human driver or operator, but require safety issues to be addressed as part of a safety management plan as an essential criterion.
4.4 Option3 - Guidelines include driver or operator duties and training requirements as essential criteria to be considered as part of a safety management plan.
4.5 Option3 - Guidelines include driver and operator fitness for duty as an essential criterion, to be considered as part of a safety management plan.
4.6 Option2 - Guidelines required a process for driving mode transition as an essential criterion, to be considered as part of a safety management plan.
4.7 Option3 - Guidelines require system failure warnings as an essential criterion, to be considered as part of a safety management plan.
Optalert believes at a minimum, drivers should be monitored for drowsiness levels in real time when in vehicles up to Level 5 of the SAE automated vehicle classifications. This will reduce the likelihood of drowsiness-related accidents and will improve public safety and confidence.
Optalert welcomes the adoption of national guidelines for automated vehicle trials and we will make ourselves available for further comment or advice.