Category: CAM

Isn’t LinkedIn marvellous? I met Lucas Noldus Ph.D., Founder & CEO of Netherlands-based Noldus Information Technology, after he liked my interview with his Global Partnership on Artificial Intelligence (GPAI) colleague, Inma Martinez.

A few messages flew back and forth, and it transpired that he’s an expert in measuring driver behaviour, particularly driver-vehicle interactions in ADAS-equipped and self-driving vehicles. That was music to my ears, so we arranged a Zoom. What follows is the highly insightful result.

LN: “The future starts here. The world is changing. We see people living longer and there are more and more interactive devices – telephones, tablets, dashboards – with which we can interact, leading to greater risk of distraction while driving. I know personally how tempting it is to use these devices, always trying to keep your eyes on the road.

“We already have fascinating developments in connected driving and now, with self-driving, the role of the driver changes significantly. That has triggered research institutes, universities, OEMs and tier one suppliers to pay more attention to the user experience for both drivers and passengers.

“All these experiences are important because how people perceive the safety and comfort will influence their buying decisions, and their recommendations to other potential users.

“For autonomous driving, how far will we go towards level five? What happens at the intermediate stages? Over the coming decades, driving tasks will gradually diminish but, until full autonomy, the driver will have to remain on standby, ready to take over in certain situations. How will the vehicle know the driver is available? How quickly can he take over? These are the topics we’re involved in as a technology company.

“We make tools to allow automotive researchers to keep the human in the loop. Traditionally, automotive research focused exclusively on improving the vehicle – better engines, drivetrains etc. Until recently, nobody paid much attention to the human being (with a brain, skeletal system, muscles, motor functions), who needs to process information through his sensory organs, draw the right conclusions and take actions.

“Now, these aspects are getting more attention, especially in relation to reduced capacity, whether due to a distracting device, drugs, alcohol or neurodegeneration. As you get older your response time becomes longer, your eyesight and hearing abilities reduce, as does the speed at which you can process information.

“These are the challenges that researchers in automotive are looking at concerning the role of the driver, now and in the future. If the automated or semi-automated system wants to give control back to the driver because its AI algorithms decide a situation is too complex, can the driver safely take over while he’s been doing something like reading or taking a nap? How many milliseconds does the brain need to be alert again?

NK: “Draft legislation seems to be proceeding on a 10-second rule, but some studies say at least 30 seconds is required.”

LN: “Situational awareness – that’s a key word in this business. Not only where am I geographically, but in what situation. Oh, I’m in a situation where the road surface is very wet, there’s a vehicle just in front of me, the exit I need is near and I’m in the wrong lane. Understanding a situation like that takes time.

“If we take a helicopter view, from our perspective as a technology company, what should be measured to understand the driver behaviour? Which sensors should we use to pick up that information? If we use a microphone, a video camera, a heartbeat monitor and a link to the ECU, how do we synchronise that?

“That’s not trivial because one sensor may be sending the sampling at 300Hz and another at 25 frames per second. That’s something my company has specialised in over the years. We’re very good at merging data from different sources, whether it’s a driving simulator continuously spitting out data, a real car, or sensors mounted in the infrastructure.

“You then need to analyse that data and pull out meaningful quantitative units that give you actionable insights. Generating large matrices is no big deal, making sense of that information is the real challenge.

“For example, in dashboard design, a manufacturer might be comparing two or three displays of road quality. A driver behaviour study with our tools will give the designer a clear answer on which design leads to the least cognitive workload, the least confusion.

“This same technical challenge can be applied to a vast number of design objectives. The vehicle manufacturer might be looking to make incremental improvements to, say, the readability of the dashboard under certain light conditions. Or they might be working on a completely new feature, like an intelligent personal in-car assistant. A number of brands are working on that, but the concept is still relatively new.

“You cannot test every scenario on the road, it’s just too dangerous, so we work with simulator manufacturers too. On the road or in the lab, we can measure a driver’s actions with eye-tracker, audio, video, face-reader and physiology in one.”

NK: “Back to LinkedIn again, I saw a post by Perry McCarthy, the F1 driver and original Stig on Top Gear, who said something like: Simulators are getting so good these days, when you make a mistake they drop three tonnes of bricks on your legs!”

LN: “You have so-called high fidelity and low fidelity simulators – the higher the fidelity, the closer you get to the real vehicle behaviour on the road, and there are all sorts of metrics to benchmark responsiveness.

“You have simple fixed-base simulators right up to motion-based simulators which can rotate, pitch and roll, move forward, backwards, sideways and up and down. For the best ones you’re talking about 10 million euros.

“We work with OEMs, tier1 suppliers, research institutes and simulator manufacturers to build-in our DriveLab software platform. We also advise on what sensors are recommended depending on what aspects of driver behaviour they want to study.

“We try to capture all the driver-vehicle interactions, so if he pushes a pedal, changes gear or turns the steering wheel, that’s all recorded and fed into the data stream. We can also record their body motion, facial expression, what they’re saying and how they’re saying it – it all tells us something about their mental state.

“Eye tracking measures the point of gaze – what your pupils are focused on. In a vehicle, that might be the left, right and rear mirrors, through the windscreen or windows, around the interior, even looking back over your shoulders. To capture all that you need multiple eye-tracking cameras. If you just want to look at, for example, how the driver perceives distance to the car in front, you can do with just two cameras rather than six.

“Eye tracking generates all sorts of data. How long the eyes have been looking at something is called dwell time. Then there’s what direction the eyes are looking in and how fast the eyes move from one fixed position to another – that’s the saccade. People doing eye tracking research measure saccades in milliseconds.

“Another important metric is pupil diameter. If the light intensity goes up, the pupil diameter decreases. Given a stable light condition, the diameter of your pupil says something about the cognitive load to your brain – the harder you have to think, the wider your pupils will open. If you’re tired, your blink rate will go up. There’s a normal natural blink rate to refresh the fluid on your eyes with a fully awake person, but if you’re falling asleep the blink rate changes. It’s a very useful instrument.

“Then there’s body worn sensors that measure physiology. It’s harder to do in-car, but in a lab people don’t mind wearing electromyography (EMG) sensors to measure muscle tension. If you’re a designer and you want to know how easy it is for an 80-year-old lady to operate a gearshift, you need to know how much muscle power she has to exert.

“We also measure the pulse rate with a technique called photoplethysmography (PPG), like in a sports watch. From the PPG signal you can derive the heart rate (HR). However, a more accurate method is an electrocardiogram (ECG), which is based on the electrical activity of the heart.

“Further still, we measure galvanic skin response (GSR), also called electrodermal activity (EDA), the level of sweating of your skin. The more nervous you get, the more you sweat. If you’re a bit late braking approaching a traffic jam, your GSR level will jump up. A few body parts are really good for capturing GSR – the wrist, palm, fingers, and the foot.

“We also measure oxygen saturation in the blood with near infrared spectroscopy (NIRS) and brain activity with an electroencephalogram (EEG). Both EEG and NIRS show which brain region is activated.

“Another incredibly useful technique is face reading. Simply by pointing a video camera at someone’s face we can plot 500 points – the surroundings of the eyebrows, the eyelids, the nose, chin, mouth, lips. We feed this into a neural network model and classify it against a database of tens of thousands of annotated images, allowing us to identify basic emotions – happy, sad, angry, surprised, disgusted, scared or neutral. You can capture that from one photograph. For other states, like boredom or confusion, you need a series of images.

“These days we can even capture the heart rate just by looking at the face – tiny changes in colour resulting from the pulsation of the blood vessels in the skin. This field of face reading is evolving every year and I dare to claim that we are leading the pack with our tool.

“Doing this in the lab is one thing, doing it in a real car is another challenge, being able to keep your focus on the driver’s face and deal with variable backgrounds. Of course, cars also drive at night so the next question is can you do all this in darkness? We turned our company van into an instrumented vehicle and my sons agreed to be the guinea pigs.

“It took some work – overcoming the issue of light striking the face and causing sharp shadows, for instance – but we can now use infrared illuminators with our FaceReader software to make these measurements in full darkness.

“The turning of the head is also very important in studying distraction, for example, if the driver looks sideways for too long, or nods their head in sleepiness. When something shocks someone, we see the face change and the blood pressure rise, and these readings are synchronised in DriveLab.

“It is well proven that even things like changing radio station can be very distracting. Taking your eyes off the road for just a few seconds is dangerous. As we move to more and more connected devices, touchscreens and voice commands, minimising distraction is vital to ensure safety.”

NK: “I absolutely love this tech but what I actually drive is a 7-year-old Suzuki Swift Sport with a petrol engine and a manual gearbox, and I quite like it that way”

LN: “I’m doing research on cars of the future with my software but I am personally driving a 30-year old soft-top Saab 900. That’s my ultimate relaxation, getting away from high tech for a moment.

“At Noldus, we’re constantly pushing the boundaries of research, working with top level organisations in automotive – Bosch, Cat, Daimler, Fiat, Honda, Isuzu, Land Rover, Mazda, Nissan, Scania, Skoda, Toyota, Valeo and Volvo, to name just a few – and also with the Netherlands Aerospace Centre (NLR) and the Maritime Research Institute Netherlands (MARIN).

“Our aim is make it so that the client doesn’t have to worry about things like hardware to software connections – we do that for them so they can focus on their research or design challenge.”

For further info see noldus.com