Erik Coelingh, Senior Technical Leader for Safety and Driver Support Technologies, explains how Volvo Cars is developing self-driving cars
What is so unique about Volvo Cars’ approach to autonomous cars?
The number one aspect is that we are not doing this for the sake of technology. We take the human as the starting point and try to understand how we can turn this into a benefit for both our customers and for society, in terms of getting a transportation system that’s safer, more environmentally sound and has less congestion.
The other aspect is that the way we do it is not by creating show-off concept cars, but by bringing the technology and its benefits to the real world as part of the Drive Me Program. It means we have to answer all the questions that come up on real roads, solve all the practical problems. We are forcing ourselves to break new ground, and that’s what nobody else is doing.
In terms of the technology, how much difference is there between the self-driving cars that are being developed by various different companies?
The components and the sub-systems are similar. The key is the integration, the writing of the software.
One powerful thing that we have that other brands don’t is that we are a comparatively small company – we don’t have a separate research organisation. We have research and product development in one organisation so we have the capability to develop an idea into a product very quickly. We’ve shown that with our active safety technologies. We are the ones that put innovative technology into production first.
Does Volvo’s leadership in safety make it easier for people to trust a Volvo autonomous car?
I think that self-driving cars and the Volvo Cars brand fit very well together. It’s because we put the person at the centre, not the technology. It’s about making sure the person is relaxed, has trust in the technology and enjoys the journey.
Can you really trust the technology?
You have to be able to trust a self-driving car, otherwise it’s no use. We have put a lot of effort into understanding how to build up this trust. We don’t have all the answers yet, but it has to do with providing the driver with the right amount of information, being able to predict what the car is about to do, and confirming that the car has detected any potential hazards. It’s also important that the car should behave smoothly and predictably.
I compare it to turbulence in an aeroplane. If you suddenly get turbulence everyone looks up and thinks ‘what’s happening?’ but if the pilot warns the passengers in advance they don’t react.
What happens if something breaks?
This is one of the fundamental things that you have to solve when you design and build a self-driving car. The solution that we are working on is a fully redundant system. That means the technology still works if any part fails. It means that for almost everything there are two components or pieces of software. If one of them fails we have another one that can take over the task. That’s true for computers, sensors, power sources.
All of these things are very unlikely to happen, but even these small risks are something that we cannot accept. We list everything that could possibly go wrong and what the consequences are. Then we design this redundancy into the system.
How does the car recognise that there’s something wrong?
The software has to monitor itself. There are many different pieces of software in the car, which can detect whether a sensor is blocked because or if an actuator is not responding, for example.
There are different computers, in different parts of the car, which check on each other. As soon as a fault is detected, the car switches to the back-up solution.
How can you deal with the unpredictability of other road users?
This is one of the questions the Drive Me project will hopefully shed some light on. Uncertainty is dealt with by always taking a safe and conservative approach. Only when the vehicle has sufficient information can it proceed to drive autonomously. When the uncertainty is too large, the vehicle may slow down or request the driver to take over again.
What will the car do when it must decide between various different dangers, be it another car, a pedestrian or leaving the road?
The key for a self-driving car is to not put itself into a situation like that. It will always keep a safe following distance, for example – at least as large as the car’s stopping distance. If it is uncertain because it can’t measure something it will slow down. A good example is that if there is a parked car at the side of the road that the sensors can’t see behind, the self-driving car will slow down or change lanes to create a safe margin.
The self-driving car will follow traffic rules. It will continue in its lane, it will brake as hard as necessary to avoid a collision, but it won’t perform a highly dynamic, unpredictable avoidance manoeuvre that could create more danger.
The cloud will be hugely important. Will every carmaker have its own cloud or will there be a communal one?
Every car brand will have its own cloud to ensure reliability, safety and also user privacy. But different clouds will interface to exchange defined and pre-agreed bits of information. There will also be official clouds run by the transport authorities. This will contain information on lane closures, emergency vehicles, weather and so on.
What happens if you lose your internet connection?
AD will only be possible when the Volvo cloud issues an approval signal. This signal could then be renewed, for example every minute. If there is no signal or a negative signal, the car would alert the driver to take over. If the driver doesn’t take over, the car would bring itself to a safe stop.
Is car-to-car communication necessary?
Autonomous cars won’t need to communicate directly with each other. If they did we couldn’t deploy self-driving cars, because most cars on the road do not have it. It would be useful but we can live without it.
How much work is needed to integrate AD cars into the transport infrastructure and society?
Self-driving cars will be deployed on the roads as we know them today. We do not require modifications to the infrastructure.
In the future, when there are many autonomous cars, it will be possible to design the infrastructure in a different way to make the most of them. But the cars must come first.
There are other societal aspects that do need to change before self-driving cars become a reality, mainly to do with traffic regulations and certifications of vehicles.
Will drivers treat autonomous cars differently?
We don’t know how real people will interact with the AD functionality, or which user interface is best. Drive Me will help us refine this – we will find out about human-machine interaction in real life.
What we want to eliminate above all is a scenario we call “mode confusion”, where the human thinks the car is driving and the car thinks the human is driving. Then you would have a car out of control – and that, of course, has to be avoided at all costs.
What about road conditions, weather and all those things you can’t control – can you allow for them?
In extreme weather conditions, for example a snow storm, autonomous driving would not be available.
Are there any similarities with aircraft autopilot systems?
We have redundant systems [duplicated or triplicated safety critical components] similar to aircraft but we also have a safe stop function. That’s different to aircraft, which are built to keep flying.
What is the biggest challenge autonomous driving has to overcome?
The biggest challenge is to make the technology so reliable that the driver can safely do something else.
More specifically, we simply don’t know what challenges might arise until we have a trial. And that’s why Drive Me is so important. On paper, we think safety will increase, fuel economy will be better and so on. But we simply haven’t put this to the test. So Drive Me will give us an idea whether the theoretical advantages will materialise as we predict.
Various companies have demonstrated autonomous cars over the past few years. How far are such cars from reaching the road?
There’s a huge difference between showing off a car as a concept and building a real product. You don’t need back-up technology for the demonstration. But if you have a real customer behind the wheel the requirements are much tougher.
When will fully autonomous cars be available to customers?
Real customers will get behind the wheel of self-driving cars for the first time in 2017 as part of the Drive Me program.
After that I expect things to move relatively slowly – self-driving cars won’t be on sale until, say, after 2020.
The reason is that we will have to prove that the self-driving car is safe in the environment in which it is used. There will be huge efforts to verify that the Drive Me cars are safe in Gothenburg traffic, but that doesn’t mean that you could take the same car and it will be safe in London. There may be exceptional situations there that the car has never encountered.
You have to perform verification activities wherever customers will drive the car because there are big variations in infrastructure, weather conditions and driving behaviours across the world.