A car drives autonomously and decides independently to change lane and to overtake. No more need for driver intervention. Science Fiction has become a reality.
The French company Valeo has long since recognized this trend: self-driving cars are going to revolutionise the automotive industry profoundly. In South Gerrmany, Valeo is testing fully automated vehicles on authorised sections of motorway. Artificial intelligence plays a key role here.
Harald Barth: “AI is in a position to a process a significantly higher volume of data – and also, very importantly, to not only measure situations but also understand them so as to be able to create the proactive element that is so essential in automated driving.”
To gain this understanding, the AI needs real data, generated on measurement vehicles with the aid of various sensors. This is because road traffic is too complex to be successfully captured by a formal set of rules. To guarantee road safety, an autonomously propelled vehicle has to be able to recognize and also classify different objects. The keys to success here: Machine Learning and Deep Learning.
Harald Barth: “Deep Neuro Net, that is, neural networks, work with a great deal of training data. In other words, a person perceives their environment and learns to understand it, and we’re giving these systems that same kind of understanding.”
In Saarbrücken the DFKI, the German Research Centre for Artificial Intelligence, is also working on the future of mobility. The scientists here are taking a different approach:
Christian Müller: “Actually our unique selling points are that we look after the vulnerable road users. Our work has less to do with motorway traffic and more with inner-city traffic. That’s where we’re modelling human behaviour by using AI.”
With a so-called “motion capture suit”, the researchers collect data on the walking behaviour of pedestrians. The data are then fed into the system on board the autonomous vehicle. The advantage here is that the car’s artificial intelligence should, by means of countless case scenarios, learn how to respond proactively to pedestrian behaviour.
Christian Müller: “While industry limits itself to merely detecting pedestrians, we’re actually predicting their behaviour. That gives us a head start of 5 to 7 seconds, during which we can predict people’s future behaviour or their future movements.”
But it’s going to be some years before vehicles can drive themselves fully automatically. There are still many challenges to be overcome – technological as well as legal.
Harald Barth: ”What I predict is that in ten to twenty years’ time, most of new cars will be equipped with systems where you can at least partially give them driving tasks.”
Christian Müller: “If we’re talking about inner-city traffic, then we’ll certainly need to discuss structural measures – because the proximity of vulnerable road users prevents driving there with any high degree of automation.”
Artificial intelligence will also revolutionise many other sectors. Find out how it can make your production more autonomous – at automatica in Munich!
Социальные закладки