There is a childhood game with very simple rules. Throw something to someone who is clearly daydreaming, shout “think fast” …see if they catch it.
Hearing these words probably meant that something had already been thrown in your direction. There is a flash of panic as you plummet back to reality. You take in your surroundings, see people staring at you, look for the object and -hopefully- catch it.
It’s not the most advanced game but, provided the object is not too fast or heavy, it is mostly harmless. Unfortunately, we’re rapidly approaching the time where our autonomous cars will be able to say “think fast” while we are daydreaming, plunging us back into the role of driver. Even more unfortunately, the objects in this scenario are very fast. Very heavy.
Vehicle manufacturers have entered a new phase. The era of Human-Machine Co-operation, where human and machine need to be aware of each other in order to deliver higher levels of automation. For the next decade at least, an autopilot system will rely on returning control to a human driver in many situations. To deliver expected levels of safety, it is vital that we have technology that robustly supports this transition, and driver monitoring is a key solution to achieve it. Where the Think Fast game relies on recognising that the subject is not prepared to catch something, the future of autonomous systems depends on vehicles being able to recognise that their driver is prepared to take control.
Driver monitoring systems support manually driven vehicles by recognising the driver’s current level of alertness and issue a warning when their level of drowsiness could compromise their ability to drive safely. For autonomous cars, the driver monitoring system must also incorporate a predictive element – is the driver going to be fit to regain control soon?
Monitoring someone’s state while they are driving is relatively easy. Their position, the relative location of their head, gaze, eyes, hands etc, are mostly consistent. However, monitoring the state of the ‘driver’ when the vehicle is being driven autonomously is far more challenging. The driver’s body position is much less predictable.
The role for future driver monitoring systems will increase as SAE Level 2 systems become more common-place, as the systems will need to ensure the driver remains alert and attentive while controlling the vehicle, ‘police’ that the self-driving system is being used correctly, and check that the driver remains alert and is monitoring traffic conditions when not in control.
The role for driver monitoring systems becomes much more challenging at SAE Level 3 and above. Now the system needs to perform all of the SAE Level 2-based tasks, and predict the driver’s ‘readiness to re-engage’, stopping whatever business or recreational task they were performing and taking back control of the vehicle.
This role is perhaps the greatest challenge of all, as driver monitoring will need to continually understand the situation both inside and outside the vehicle and decide what action to take in real time. Once the machine is engaged and driving, the key role of driver monitoring is to ensure that the driver is ready and capable of taking back control within the warning timeframe. This amount of time will continually vary dependent on multiple factors, including vehicle speed, traffic conditions, road type, and weather.
To be effective, the system also needs to understand the condition of the driver. Each person has a unique optimum stress level at which they deliver optimal performance. So, the optimum point of reaction for each person is a function of task-load - too much and they may freeze, too little and they may relax and become drowsy.
The system has an almost impossible task - to guesstimate a driver’s stress (and/or cognitive workload) and use this information to adjust the take-over warning interval. This task is so challenging that not even the rail or air industry have yet resolved it. This is new territory, but is a vital part of any robust solution for automated driving systems needing to default back to a human driver.
As SAE Level 2 vehicles continue to be more commonplace, we are seeing a much greater need for effective driver monitoring. Recently, a ‘driver’ of a Tesla Model S was asleep while the car was driving on the motorway. While officers in a passing police car realised and were able to wake him (and confiscate his license), at no point did the vehicle know it was operating unsupervised.
“Driver monitoring is not a legal requirement yet, but it is the key to a safer Level 2. For Level 3, it is critical” says Dr Alain Dunoyer, Head of Autonomous at SBD Automotive. “Until more large-scale studies have been completed, OEMs should be very careful, constraining the envelope of operation of their autonomous systems with technologies like driver monitoring, ensuring they are used in the right situations.”