With just a tap of the turn signal, Daimler’s self-driving S-Class changes lanes in dense highway traffic. The specially trained driver keeps his hands on the wheel, just in case. But he doesn’t need to use his feet at all.
The car keeps a safe distance from other vehicles, and accelerates and brakes automatically. What's more, the driver needn't worry about traffic tickets either, since the car can be programmed to obey speed limits at all times.
Despite some impressive feats, the S-Class's test drive near the German town of Sindelfingen also showed that the automated car has plenty to learn. It still can’t detect traffic signals, while the driver has to take over braking when entering a roundabout with oncoming traffic from the left.
These extra features would bridge the gap to let drivers feel entirely comfortable in an automated vehicle. For its part, Audi is conducting field tests to link its cars to traffic signal systems in major US cities – but it still needs to access the necessary data.
Even so, the technology is making some serious progress, thanks in large part to new artificial intelligence processes. Daimler engineers, for example, have fed thousands of images from German cities into their software in order to train the system how to handle complex traffic situations. As such, the S-Class’s automatic recognition of speed limits and traffic signs works surprisingly well.
Artificial Intelligence (AI) in cars is common throughout the entire automobile industry, and high-ranking industry officials discussed the topic at length at a recent conference hosted by the German magazine Auto Motorsport.
Autoparts supplier Bosch announced its desire to develop a new “brain” for cars for the coming decade, specifically an on-board computer that will interpret traffic situations and predict other drivers’ behaviour.
Bosch competitor Continental is also working on a research partnership with Oxford University to investigate which algorithms can improve a car’s optical object recognition. Meanwhile, the American computer chip maker Nvidia is working on technologies that not only allow cars to recognise faces, but also read lips.
Juergen Schmidhuber, head of the Swiss Institute for Artificial Intelligence (IDSIA), sees an electronic reproduction of the human brain through electronic networking in the not-so-distant future. Nevertheless, “we still need another 25 years to become competitive,” he admits.
In any case, it will still be a while before computers can take full control in cars. As a first step, the vehicles are likely to move in highly restricted spaces, such as car parks without drivers.
When it comes to secondary tasks not related to the driving itself – such as when offering navigation suggestions or predicting musical selections – artificial intelligence is already in use, says Frederik Diederichs of the Fraunhofer Institute for Industrial Engineering in Stuttgart. "However," he notes, "security-relevant decisions are not yet being taken by AI systems."
On the one hand, the legal basis for AI decision-making is still lacking in many jurisdictions. At the same time, another problem is that AI-based decisions aren’t based on data that corresponds to human perception.
"The decision-making processes are unpredictable, in that they are based strictly on probabilities," says Diederichs. However, the key issues here is that drivers can always understand and explain why a specific decision was made. According to Diederichs, "with AI systems, this justification is quickly lost."
At Bosch, programmers can double-check the possible range of decisions that a car will make. "The leading manufacturers tend to err on the side of safety," says Michael Frey of the Karlsruhe Institute of Technology. Up until now, however, there has been an absence of standardized test methods for artificial intelligence in cars.
At the Fraunhofer Institute’s so-called Neurolab, and alongside her colleague Diederichs, Kathrin Pollmann is investigating the effects that intelligent systems have on users – and is taking her research a step further.
In this case, a key research question is, for example, how a driver’s concentration levels can be taken into account, such that he can intervene at any time if necessary. Pollmann also wants to measure the extent to which drivers’ decisions correspond to those of the automated assistance systems.
Although automakers are including an increasing amount of automated assistance systems in their vehicles, there is still a high degree of generalized scepticism.
For example, the consulting company Deloitte determined that, even for those technologies with a safety certificate, only 47% of German consumers would drive with an autopilot. In China, however, that figure rises to an impressive 81%.
In the end, systems that don’t stop at a red traffic light are unlikely to increase public confidence. As such, for the near future at least, the main comfort highlight for the S-Class will remain the warming massage function now featured in the rear seat. — dpa