Self-driving cars will save us from ourselves. Statistics show that they would reduce road accidents by up to 94% because crashes are usually caused by driver error. Yet it is humanity’s lack of trust and acceptance of them that is part of the problem in achieving a world full of autonomous vehicles (AVs). Designing AVs to make passengers feel comfortable and safe is therefore a critical step in crossing the chasm between level 3 (where the driver must be ready to take over when needed) and level 4 autonomy (where the driver can go to sleep).
AI in the driving seat
Innovative design often takes inspiration from seemingly unrelated areas of science. In that spirit, I pose this question to help us respond to the challenge: can we use the principle behind why we can’t tickle ourselves to design better AVs?
Since humans have existed there has always been an inertia in the mass population about accepting new, ground-breaking technology. Victorians initially believed that travelling on railway carriages caused insanity, and in the early 1900s telephones were thought to attract evil spirits – or at least thunderstorms. When these technologies were accepted as being safe and reliable, they became more widespread and went on to change the world. If AVs were designed with the comfort and behaviour of passengers in mind, unlike Tesla’s recent AV updates, they would readily be considered a help rather than a hindrance.
Why can’t you tickle yourself?
Imagine a situation where you are travelling in your AV enjoying a cup of hot coffee and reading the news on your virtual headset. Suddenly the car screeches to a halt for a few seconds before accelerating back to normal speed again. This is likely to cause you a lot of discomfort, and you are unlikely to shout about the greatness of self-driving cars to your friends. Unbeknownst to you the AV was acting in the interest of your safety as there was a cyclist who had suddenly swerved into your path. Ah, the perils of living in Cambridge, UK!
If you had known what the car was going to do, your body would have been prepared for the sudden deceleration, and your muscles would have reacted in a way that minimised discomfort. The reasoning behind this is the same as is behind why we cannot tickle ourselves. Go ahead, try it! Your brain uses the Sensory Cancellation Principle where an internal model of your body is used to predict how your muscles will react to an input (the tickle stimulus). If you tickle yourself the input is known exactly, and the internal model’s predicted reaction matches your body’s actual reaction. If someone else tickles you, the input is unknown and the internal model’s predicted reaction will differ from what your muscles actually do. It is this difference between prediction and reality which becomes ticklishness or discomfort.
Designing for comfort
But how do we quantify discomfort for a passenger in an autonomous vehicle? If we are to consider simply the vehicle’s acceleration or braking, the angle between the passenger’s head and neck can be taken as a useful discomfort metric. The greater this angle, the harder the passenger’s neck muscles would need to work in response to avoid injury. Discomfort – usually a subjective concept – can be mathematically represented in many more ways depending on the type of vehicle motion that might cause it.
Autonomous vehicles should therefore be designed to move in a way that minimises passenger discomfort, and the way in which the human body controls its muscles should be taken into consideration. One way in which some level of autonomy is already being realised is in active cruise control. Consider our case again where an AV must suddenly slow down due to some obstruction ahead. Once the path ahead becomes clear again, the vehicle should resume its original speed. Accelerate too slowly and time is wasted but accelerate suddenly (‘warp speed’) and the passenger feels a strong jerk. How then, should our vehicle speed up?
We can simulate the effects of various acceleration profiles on the passenger’s head and neck to predict the discomfort caused. This would save time and money currently spent in testing. In reality the passenger may have a vague idea of how the car is going to move, and the simulation could be improved by allowing the passenger model to ‘see into the future’. Research has shown that as in reality, knowing the vehicle’s future motion would reduce the discomfort felt.
Increasing passenger trust
How would we ensure the passenger knew about the upcoming motion? In-vehicle features such as audible and visual prompts could help to enable the passenger to anticipate new movement. The use of lighting, spoken announcements and visual displays are just some of the potential indicators of upcoming movement that could be used. These design features would contribute towards achieving the ultimate aim of increasing passenger trust in autonomous vehicles, as the car would not seem to behave unexpectedly.
It is not just passengers who need to feel safe in and trust the AV, it is pedestrians on the streets too and this should also be addressed in the design stage. Current vehicles already have turn indicators – this feature could be extended to indicate the speed of the vehicle too. A natural consequence of the safety-conscious environment for autonomous vehicle design is that these AVs will be prone to pranks. It would be very easy to repeatedly step into the path of an oncoming AV, knowing that it would always have to stop or slow down. Exterior information on the car’s speed may perhaps deter some pranksters. Not enough thought is currently given to how the general public might interact with AVs, but at Cambridge Consultants we are already developing technology that integrates AV design with human behaviour. We have created an AV simulator to run high throughput validation of perception, planning, prediction and control algorithm designs.
The road ahead
Science fiction often presents us with a futuristic world where all transportation is autonomous, but we must be realistic in imagining the path taken to reach this future. There is a large gap between level 3 and level 4 autonomy, partially due to technological barriers but mostly due to regulatory concerns and public acceptance. It’s likely then, that a large-scale deployment of autonomous vehicles is expected to be seen in applications with minimal human interaction, such as in rural areas or farming.
In these situations, the importance of simulating human discomfort would still be useful as the motion of the AV may be more unpredictable due to rough terrain or unstable weather. A large role would need to be played by signal processing techniques (including the use of Kalman filters which deal well with uncertainties) and predictive control theory in vehicle design and simulation – areas where Cambridge Consultants has a great deal of expertise.
There is no doubt that a fully autonomous future would benefit society greatly through safer roads and increased human productivity. Just think of all the things you could do on your commute to work! But there is a long way to go until that idyllic era; many technological advancements need to be made first, keeping human interaction in the design loop. I’d like to invite you to contact us and be a part of that change. Work with us to create breakthroughs that connect many areas of science and ask the right – and sometimes strange – questions. Can a self-driving car tickle you?