Prof. Dr Ernst Dieter Dickmanns is regarded as the pioneer of autonomous “seeing” cars. His “computer vision” methodology developed in the 1980s is still in use today in autonomous vehicles.
Prof. Ernst Dieter Dickmanns is more than a little envious when he sees the technologies that autonomous vehicle developers have at their disposal nowadays. “Computing power per microprocessor today is almost a million times greater than when we started. The volumes of computers and sensors are less than a thousandth of what they were back then.” “Back then” was the late 1980s, when Prof. Dickmanns, born in 1936, began developing an autonomous car. Right from the start, he worked on what is now commonly termed “computer vision”. “When you look at the role vision plays in biological systems, it has to offer major advantages for technical systems too.” That was the idea that led him to develop a method of teaching cars to see.
Seeing in real time, even without high computing power
“Even back in 1975, we were seeing computing power per microprocessor increasing by a factor of 10 every four to five years,” Prof. Dickmanns recalls. 1975 was the year when he – at the age of just under 40 – joined the University of the German Armed Forces in Munich. “It was likely that computing power would increase by a factor of a million by the time I retired. That was likely to be enough to permit video analysis in real time, which would be an entirely new technical accomplishment.” Dickmanns and his team began developing a method that was to turn “computer vision” into a reality before his retirement. In the 4D method, as he terms it, the data captured by cameras is digitised and processed by the computer merely as abstract lines with adjacent grey-scale areas. Rather than comparing the current image against the previous one, as was common practice at the time, he used motion models in three dimensions and integrated time (which is why the method is termed “4D”) in order to understand the observed process in the real world. These models predicted the expected features in the next image. As a result, much less data was created. Even using the processors available at the time, simplified scenes could be processed in 100 milliseconds – corresponding to real time in the automation world.
“I do consider safe autonomous driving on all kinds of high speed roads to be important.”
From Munich to Copenhagen – almost fully autonomously
The first vehicle fitted out in this way was running autonomously on blocked-off test routes as far back as 1987. The technical systems needed to do that literally filled cabinets – the first test vehicle was a Mercedes van with a five-ton payload: the VaMoRs (a German acronym standing for “test vehicle for autonomous mobility and computer vision”) provided sufficient capacity to accommodate a power generator and several metres of industrial control cabinets for the electronics. But just a few years later, the space taken up by the equipment had shrunk significantly. The “VaMP” vehicles from the University of the German Armed Forces and ViTA-2 from Daimler were both based on a Mercedes saloon car, and were results of the Prometheus project initiated by the European automotive industry. Both projects were supervised by Prof. Dickmanns. From 1993 onwards, the cars were able to run completely autonomously on roads with normal traffic. The crowning glory was a trip from Munich to Copenhagen. The test vehicle covered 95 per cent of the 1,700-kilometre journey with no intervention by the driver – changing lanes, overtaking other vehicles, and attaining a top speed of 175 kilometres per hour. It was a sensation back then – and remains a benchmark for modern-day self-driving cars. The 4D methodology is now an integral element of autonomous vehicles, and Prof. Ernst Dieter Dickmanns is acknowledged globally as the pioneer who taught cars to see.
Recipe for successful research
By the time he retired in 2001, he had developed many more solutions in the fields of computer vision and autonomous driving. Asked what his recipe for success might be, he lists four points: “The conviction that your idea is better than any other; acquiring adequate research funding; selecting appropriately qualified staff and doctoral students; and engaging widely in intensive dialogue with partners in industry as well as with international scientist colleagues at conferences.” He believes that anyone capable of bringing those attributes to bear on a project, alongside their own creativity, has what it takes to be a successful researcher and inventor. Not everything Dickmanns experienced in his scientific career was positive, however. Looking back, he offers a word of advice: “I would be even more cautious than I was in selecting industrial and scientific research partners, and make sure that key points are laid down in writing.”
No worries
But Prof. Dickmanns is far from being truly retired. He continues to follow developments in autonomous driving, gives lectures, and still has ideas about how to improve computer vision: “Given the current state of the art in technology, I would directly target solutions that have proven their worth in biological systems. One of the features I would install would be little eyes with dynamic direction-of-view stabilisation and control, fitted at the top of the A-pillar on the left- and right-hand sides of the car.”
Would he really be prepared to ride in a fully autonomous car today? His answer is clear: “If there was one, yes.” He can also imagine buying one of the new cars that are able to park themselves and drive autonomously in traffic jams: “Though I don’t place great value on parking. I do, however, consider safe autonomous driving on all kinds of high-speed roads to be important.
