 Safety 2
They say paper is patient; the Internet, perhaps, even more so. There are so-called compelling arguments whose foundation is questionable. They are primarily plausible, which is mistakenly
confused with
'true'. Here's an example: If the 'human' factor plays a role in 93 percent of all accidents, then autonomous driving can prevent 93 percent of all accidents.
At the Rheinisch-Westfälischen Technischen Hochschule of Aachen, this claim has been investigated and critically examined. It's important
to note that there is no hotbed of opponents of progress there. On the contrary, years ago people were already speeding down the motoway A4 in specially modified test cars with special
permits, taking their hands off the steering wheel, or, rather, letting the cars find parking spots on their own.
By the way, the latter would also be a solution for our parking spaces, which are often far too narrow, especially if cars keep getting wider, because with self-parking cars, ultimately no one
needs to get out or get in. But that's just a side note, and it could certainly become the norm before long.
Yes, even self-driving cars are not accident-free, as demonstrated by the fatal collision between an Uber vehicle and a woman who was pushing a bicycle across the street. The vehicle was
reportedly capable of Level 4 autonomy, but apparently failed to recognize the woman on the bicycle as such. The driver is said to have been using his smartphone.
And although in another case of a collision between a skidding car and a vehicle from the Google fleet the latter is not at fault, it is still facing a barrage of criticism. Because in the photo, the
road is strangely swept clean. It seems as though one of the drivers made an effort to somehow avoid hitting the other. Apparently, this isn't yet sufficiently integrated into Google's software. At least
no one was hurt.
You may already suspect that autonomous driving, of course, could also give rise to its own category of accidents. That certainly undermines the 'Zero Accidents' theory. But these are just
speculations for now. The people at the Technical University have delved much deeper into the matter. To prepare for a major research paper, they first looked at accident statistics from
previous years.
One significant example was the seat belt requirement introduced in Germany, which has had a clearly visible impact on the number of traffic accident fatalities. Even more spectacular was the
failed moose test shortly before the turn of the millennium, which led Mercedes to introduce ESP in all its vehicles. Here, too, the statistical evidence is clear.
But they were faced with a dilemma. Practical experiments failed to validate the 'accident-free period' hypothesis. There were only Level 2 autonomous vehicles, such as the Tesla, for which
restrictions were even imposed following an accident. Level 3 may have been promised for the Audi A8, but it hasn't been released yet. By the way, a vehicle like this, which has been
customized solely for research purposes, can cost between half a million and one million euros.
As a result, an impact analysis was conducted. Part of this involves drawing on existing statistics. For example, since 2006 there has been the Insurers' Accident Research Initiative and the
Geographic Information Systems Database, which contain much more information than just the location and time of the accident.
Die Wissenschaft versucht Ordnung bzw. Struktur zu schaffen. For example, the accidents are sorted according to how they occurred and even have a number. So if an accident researcher mentions
number 646, the others know that the accident classified as such occurred when a vehicle was merging from the left. This is how accident types are created.
It is also important to note that, when reevaluating these types of accidents, the analysis assumed a mix of human-driven and automated vehicles. The findings showed that such accidents will continue
to occur, albeit less frequently. Automated traffic simply lacks the spontaneous behavior that often leads to accidents.
The overall volume of traffic makes analysis difficult. Even a review of today's accident data suggests that it makes sense to divide traffic into at least three categories: city, rural, and highway traffic.
The latter was singled out, and self-driving cars were 'simulated' into it at various percentages. What was important was not only relevant driving scenes, but also the information provided by the
manufacturers regarding how such cars behave or react based on the current state of the art.
Such a vehicle could, for example, 'see' 200 meters ahead and behind it, with a wider cone of vision extending 80 meters ahead. That came out to only 10 meters on each side. The driving scenes were
automatically derived from videos recorded over a period of half a year during trips made by a total of 100 vehicles. It is also worth noting that accidents on highways account for only 6 percent of all
accidents resulting in personal injury. Most accidents occur, for example, at intersections in urban areas.
The results were sobering: With 5 percent autonomy, only 4 percent of accidents are prevented; with 25 percent, the figure is 17 percent; and with 50 percent, the reduction rate is 30 percent. It even
reduces the total distance traveled on the highway by 53 percent. However, commercial vehicles were left out of the equation.
And, of course, all conditions that increase the risk of accidents, such as black ice or fog, which are also quite challenging for the sensors to handle, were left out. Particular attention is drawn to a rare
case: the deliberate disruption of motorway traffic, an act that borders on suicide and consciously accepts the risk of other victims. Can this be avoided through automation?
As you can see, not much remains of that seemingly plausible prediction of a reduction in accidents. Not even, and especially not, when one considers that full automation of the highway will likely not
be feasible, at least in the foreseeable future. The TH employee's forecast was very favorable for the many little helpers, but not very favorable for autonomous driving. The term 'decades' came up.
But so as not to end on such a destructive note, here's the latest project that RWTH is proud of because it's being carried out under its leadership: 'L3Pilot - ika'. 1,000 drivers are operating a total of 100
Level 3 vehicles from 13 manufacturers across 10 European countries to conduct a cost-benefit analysis. For more information, see: l3pilot.eu.
And here’s something else that was already noted in this book in the context of artificial intelligence: Research on automated driving seems to show that human drivers, provided they are attentive, are
apparently so skilled at controlling the vehicle that digital electronics cannot really replace them easily.
| 
