This year, I have spoken with key decision makers in nearly every major automaker and driver monitoring system (DMS) supplier. Within our team at Optalert, we have a very clear position on driver monitoring: If a driver is objectively impaired (whether by drowsiness, intoxication, or anything else), there must be an intervention. If they are not impaired, there should be no intervention.
An argument I sometimes encounter is that safety systems should also factor in the driver’s subjective self-assessment to increase consumer acceptance. The idea is that “consumers will accept a system that beeps at them when they expect it to.” In this piece, I deconstruct the problems with this position.
Below I break down the two scenarios in which a decision needs to be made one way or the other, using impairment from drowsiness as an example. But first, I share an overview of the two aspects of drowsiness and how well they correlate.
If this is familiar to you, jump down to the section: “Two cases in which tiredness and impairment differ”.
Drowsiness is expressed biologically in two ways.
Tiredness is the subjective feeling of needing to rest or sleep. It involves a person gauging how tired or “sleepy” they feel. The dominant measure for subjective tiredness is the Karolinska Sleepiness Scale (KSS).
Impairment is the increase in relative risk of performance failures. As a driver becomes increasingly impaired from drowsiness, their relative risk increases. In a laboratory-based psychomotor vigilance task, this is measured by errors of omission. In driving tasks, it is measured by errors such as wheels drifting out of lane. Optalert’s Johns Drowsiness Scale (JDS) uses the objective biomarker of eyelid movements to quantify impairment. All of these are objective measurements that are independently observed. They are objective because they do not involve subjective judgement either by the subject under observation or anyone else observing them.
Subjective tiredness as measured by the KSS has exhibited a moderate correlation to impairment at late stages of drowsiness (KSS 7 to 9). However, it is not useful at early-stage drowsiness. It’s virtually a dice roll from 1 to 6.
An independent study kept subjects awake for 32-34 hours before a two-hour drive on a closed loop track. Performance failure was defined as a lane departure. With a threshold of either 8 or 9, KSS was incorrect 12% of the time. Optalert’s JDS (objective measure of impairment based on eyelid movements) was incorrect 4% of the time.
In a real-world driving situation, there is no issue if the driver’s self-assessment is correct: an impaired person believes they are impaired, or an unimpaired person believes they are unimpaired. A system based on objective impairment (such as the JDS) should detect the same thing as a system developed with subjective tiredness (such as KSS) as its ground truth.
There would be two possible scenarios in which subjective tiredness differs from impairment:
This is quite a common scenario we have seen a lot in our testing and in the field. Beyond a certain period of extended wakefulness, a person’s KSS may stay at 8 or 9. Anytime you enquire they will say they are “sleepy” or “very sleepy”, and “exerting effort to keep awake”. But in a lot of instances, their JDS is well below 4.5 and they can safely operate a vehicle without error.
What should the car do? First, it is very likely that the driver will have an elevated JDS above 2 or 3 in this state. If so, a DMS with an impairment-based measure could deploy soft countermeasures that prolong wakefulness and keep the driver alert. These include lowering the cabin temperature, increasing the volume of the sound system, or encouraging the driver to phone somebody. External stimulation counteracts drowsiness, especially something that demands engagement from the driver, such as conversation. The loud alert should be held in reserve to warn the driver when they are approaching a dangerous level of impairment.
In the exceptionally rare case that the driver is very alert but feels subjectively tired, the car should do nothing. Beeping at the person solely to increase trust in a DMS is an unnecessary distraction. An Esurance survey found almost a third of drivers found their car’s warning sounds distracting, prompting a quarter of drivers to deactivate at least one feature. If the end result is an impairment detection feature being deactivated, on-road safety has been compromised.
In this scenario, the driver subjectively believes they are unimpaired (KSS ≤ 6) and says, “I’m not sleepy.” However, they are objectively impaired (JDS ≥ 4.5) and their risk of an accident is at least five times that of when they are fully alert. An analogous scenario would be an intoxicated friend telling you they feel they can safely drive home.
At this point, there is a moral imperative to intervene and alert the driver with a loud warning that rouses them. Note that this will only temporarily counterbalance their drowsiness for a few minutes. They should immediately find a suitable place to stop and rest for the safety of themselves and other road users.
It is important to note that almost all the 16% to 24% of vehicle accidents caused by drowsiness involve this scenario: Someone mistakenly thought they were not too drowsy to drive. Any drowsiness detection technology must respond in this situation for maximum risk reduction. No consideration for a consumer’s acceptance of an alert should factor in when their life and that of other road users is at risk. Intervention is necessary.
Many DMS providers have rushed technologies to market in response to the European General Safety Regulation 2021/1341 mandating drowsiness detection. There is also a scramble from many engineering teams to detect driver intoxication. Consequently, there is an increasing issue with false alerts in new vehicles annoying consumers. This has led engineers with less accurate systems to lower their thresholds or attempt to align with the driver’s self-assessment in the hopes that they will align with consumer expectations for “when the system should beep at them”. It is a byproduct of the fact these systems cannot accurately measure impairment.
Across all systems that quantify driver impairment – whether drowsiness, intoxication, or something else – the driver’s self-assessment should not influence the assessment. The path to maximising trust in these systems is to ensure they accurately measure impairment and only impairment.
Our research began in 1994. We spent 14 years getting the science right before commercialising our technology in 2008. We have since spent over 15 years optimising precisely how to detect driver drowsiness from an objective biomarker (eyelid movements). No other technology has come close to the predictive power of the JDS and its ability to prevent accidents.
We encourage automotive OEMs and tier 1s to contact us for more information on how to integrate our drowsiness detection system within your DMS.
We are eager to support the industry to employ sound science and ensure we are doing all we can to keep drivers and other road users safe. Please contact us if you have any questions or suggestions.