THE SILVER VW PASSAT seems to be driving by itself. Henrik Matzke may be sitting in the driver’s seat, but his hands are in his lap, not on the steering wheel. Yet his frowning forehead and his tightly pressed lips indicate mental work – and that’s just it: Matkze is steering the car by the force of his thoughts. Within the framework of the Brain Driver project, and without as much as raising a finger, the science assistant from the AutoNOMOS innovation laboratory at the Freie Universität (FU) Berlin steers the Passat precisely around curves, accelerates as he thinks fit, and brakes without raising his feet even a millimetre from the mat. Matzke is part of the research group around Raúl Rojas, who teaches artificial intelligence at the FU Berlin. Prior to this, the scientists steered their autonomous car by iPhone, iPad and eye blink. Now it’s the turn of brain waves. Academic gimmickry? Rojas denies this. “Our long-term objective is a car that drives totally autonomously,” the professor says. And on the way to that car, his task is to fathom what is technically feasible.
Even if the new project sounds like pure science fiction, it is indicative of the potential this field has for scientists. Without human intuition, however, artificial intelligence behind the wheel is a risky business. Which is why the driver can also interrupt automatic steering by tipping the pedal or gripping the steering wheel and thus personally intervening. Even though a spectacular milestone has been reached with “thought transmission” – it is still a long way to the car that drives itself. But artificial intelligence researcher Rojas is still convinced that the vision of autonomous driving can only be realized “on the hidden path” via improved driver assistant systems (DAS).
“Thought transmission” by Brain Driver Matzke is the result of hard training at the computer. He studied the commands “left”, “right”, “accelerate” and “brake” with a sensor cap on his head that registered his brain waves and used them to produce an electroencephalogram (EEG). In keeping with the bioelectric wave patterns a dice on the computer screen moved in the corresponding direction. Only when the computer could reliably interpret and implement the wave patterns as clear orders was Matzke able to take his seat behind the steering wheel. Now the challenge was to drive more far-sightedly than in a normal car: for example, to turn a corner, Matzke had to think about this five to six seconds earlier. That is the time it takes for the thought commands to be put into practice by his sensor cap through a computer. Completely autonomous driving is actually the easier task for the researchers, as they and their Passat have already learnt that – in principle. The vehicle is equipped from roof to floor with sensors, lasers, radar and a GPS version that is precise down to a distance of 20 centimetres. Video cameras “watch” the road markings and traffic lights, laser scanners rotating 360 degrees recognize cars up to a distance of 100 metres, and radar systems at the front and rear estimate the speed and distance of other cars on the road. The lasers also provide information on the size and shape of other road users.
Yet the computer is far from being able to compete with the human brain when it comes to reacting adequately to the flood of data from this arsenal of sensors. “The art is not to gather the data, but to process it quickly and come to the right conclusions,” Rojas explains. Despite numerous test drives since 2007 – “accident-free”, as the Mexican professor emphasizes – Rojas has to admit that, “everything that seems easy for the driver is difficult for the computer.” For example, analyzing the risk of whether the slight swerve by the car in front means an abrupt turn, whether a pedestrian is about to walk carelessly onto the road or whether children at the roadside represent a danger. “Passers-by are a major problem,” says the researcher. Given that artificial intelligence always has to capitulate in the face of their spontaneity when it comes to traffic regulations, automatic drives through the city may be technically possible, but not controllable.
However, this does not prevent researchers confronting the challenges of this field. In late 2010, Google researchers under the direction of the German-born Stanford professor Sebastian Thrun managed to have a fleet of six cars drive more than 220,000 kilometres independently on national roads. And on this side of the Atlantic, researchers from the TU Braunschweig’s Chair of Cybernetics and the German Aerospace Center (DLR) even dared to tackle dense urban traffic with their VW Passat Variant, which is called Leonie. Leonie would certainly have been out of her depth with the complexities of rush-hour traffic, and she does not understand whether the traffic lights are red or green – a co-driver has to press a red or green button in the system in this case – but if the light signs were to communicate directly with the car, that problem would be solved. In fact, the DLR, among others, is working on this. It is equipping parts of Braunschweig’s Ring District with measuring instruments and communications units within the AIM project (Application Platform Intelligent Mobility). “In future, the traffic lights will be able to inform the test vehicle when they are about to change,” says Frank Köster, head of the DLR Automotive Department at the Institute of Traffic System Technology in Braunschweig. Without such communication technologies, which enable vehicles to communicate with one another, car-to-car, and with the traffic infrastructure, car-to-infrastructure, the vision of autonomic driving will not be realized. “This networking is of major importance for car intelligence,” says Köster. “Only when cars are in a position to act logically, flexibly and communicatively, will they comply with the understanding of intelligent systems.” There are also difficulties with the sensors on board. “Cameras are still not good enough and lasers are expensive,” says Bernhard Rumpe, professor of software engineering at the RWTH Aachen University. The technology is not robust enough because lasers do not like dust, whereas cameras, by contrast, fail in diffuse light, fog and darkness. Then there is the fact that a sensor-equipped car like the FU Berlin’s Passat costs roughly 250,000 euros. Sensor fusion, which involves the combination and evaluation of all the sensory data, is also difficult.
Car manufacturers particularly dread the liability issues that may arise when it comes to clarifying whether the system or the driver caused an accident. Lutz Eckstein, head of the Institute of Motor Vehicles (ika) at RWTH Aachen University, points to a fundamental legal problem: “Fully-automated driving
contradicts the Viennese Convention, which defines the human’s ultimate responsibility for driving a vehicle.” This law must be changed if a new era is to be able to begin. However, normal road traffic regulations also present an obstacle. The computers in autonomous cars have trouble learning the rules of the road. Currently, rule-based software is used in which typical situations are dealt with according to fixed rules – a procedure that is not sufficient for the special cases that occur in the road traffic regulations. So it is clear for Rojas that “as long as the entire surroundings of a car are not consistently recognized and the traffic regulations followed, these systems can only be assistants.” Raúl Rojas sees another scenario for the near future: “In terms of car-sharing, I would order a car by mobile phone and it would come to me autonomously and find itself a parking space.” The inconvenience of picking up and dropping off a car would be eliminated, sharing would be more attractive, cars used more efficiently and on the whole few of them would be on the road. The system that could well awaken a new delight in car-sharing already has a name: FASCar II, a DLR test-car that can be summoned from the parking lot by smartphone. The future of the car may possibly lie in such luxury functionality. After all, even if it were technically and legally possible to drive autonomously, it is questionable whether drivers would really sit back at their ease and not constantly question whether their car was really doing everything right. So perhaps the computer as co-pilot is the better option.////
