Selbstfahrende Autos

Brad Templeton über die Presseberichte, ein „Winter“ sei gekommen für die Entwicklung selbstfahrender Autos, welche nur zu gern von den etablierten Automarken der Presse gegenüber unterfüttert werden, um den kommenden Wandel vielleicht doch noch etwas hinauszögern zu können:

We continue to see lots of reports of a „pull back“ on robocars, a „winter“ of sorts. It is often presented as „the problem was harder than people expected“ […]

There is one factor not being talked about in the idea of a pull-back. Namely that the big car companies would not mind a slowdown at all, and so will happily confirm this narrative, whether it is true or not. The reality is these companies were dragged into this. No industry relishes the arrival of a new technology which will turn it upside down. Those are dangerous waters, and while some car companies will thrive, several will fall. Every sane OEM wants to continue the world where they were making good money selling the same cars they always made to customers, making incremental improvements.

Elektromobilität ist dank Tesla in der breiteren Öffentlichkeit angekommen, selbstfahrende Autos dank Google. Die klassischen Automarken waren hier nicht die initialen Treiber (und sind es bis heute nicht):

The arrival of electric was forced by Tesla, and the arrival of Robocars by Google. I remember talking to the robocar teams at car OEMs in the early part of this decade. Those teams had existed for some time, but they had modest budgets and were considered far-future R&D. Then Google came out of stealth, and the press were writing „The future of the car is coming from a search engine company.“ Car OEMs did not like that one bit. So suddenly the small teams became big teams with big budgets. Major efforts took place to sell the message that car OEMs were not irrelevant, that the robocar of the future would come from them, because Silicon Valley doesn’t know how to make cars.

(It’s true, that Silicon Valley still has much to learn about making cars. But it’s a mistake to think it can’t learn, or that it can’t simply pay car manufacturers to contract manufacture cars. The old knowledge of how to make cars is deep, but it’s old and thus for sale at reasonable prices.)

​Templeton über einen Teil der vielzähligen wahren Gründe, warum selbstfahrende Autos auf sich warten lassen:

The small amount of pull-back that is not about public perception comes from a few factors.

• We’re notoriously bad at estimating complex software timelines. We’re bad even when we take into account how bad we are. Even the best project managers look at their realistic timeline for a project, which might suggest it will take 2 to 5 years, and talk about how it can happen in 2 years. It might, but that’s not the most likely outcome. In this case, the range is large — 2 to 10 years, making this worse.
• For those with less experience, there is misunderstanding of the 90/10 rule which in robocars is the 99/1 rule. The first 99% takes 1% of the time. The last 1% takes 99% of the time. It’s „easy“ to produce a car that can drive a nice demo loop and stay in its lanes or drive the highway 99% or even 99.9% of the time. Sweating the details is hard, which is why Waymo is still going slow after 10 million miles of testing.
  There are legitimately people who did not understand how hard it really is.
• The talk of „Level“ 5 as a quasi-real thing blinded some to the reality that these vehicles will only operate in certain roads and driving situations for a long time. It seems that some people thought they were promised this „level“ some time in the early 2020s, which was hopefully never promised.
• The Uber fatality has raised public fear and with it developer caution. Everybody knew an injury accident would happen some day. An early fatality due to really bad practices was less anticipated, and has, as would be expected, used up a fair bit of the public’s tolerance for error.
• We’re seeing expected blowback from the public as it gets first exposure to real use of these cars. In prototype form — and in Uber’s case hurting people — you expect the public to react in a way they did not when it was all hypothetical.

​Wie langsam oder schnell selbstfahrende Autos alltägliche Realität werden, werden wir besser abschätzen können, wenn die Zahl der Pilotprojekte nach oben geht. Das wird die Jahre 2019 und 2020 bestimmen.

The best teams will be moving into pilot projects this year and the next, with other teams coming later. Once the pilot projects are underway, we’ll learn more about where the remaining issues are and how the public reacts to the technology. While people like to throw out dates, the real release date has always been „when a team can convince its lawyers and board that the technology is safe enough to deploy without creating massive risk for the company.“ That’s a milestone, not a date.

​Es ist wahrscheinlich, dass aus den Erfahrungen der Pilotprojekte auch Erkenntnisse erwachsen, was bei Straßenbau und -instandhaltung nicht nur regulatorisch sondern auch sensorenseitig der Verbreitung selbstfahrender Autos zuträglich sein könnte.

Denn es gibt Regionen international, vor allem asiatische Metropolregionen, die lieber heute als morgen ihren Verkehr mit selbstfahrenden Autos optimieren und damit entlasten wollen. Dafür werden dann dort auch Gelder für Infrastrukturprojekte bereitstehen.

The Economist über die plausibelste Entwicklung selbstfahrender Autos. Zuerst ergibt nur ein Sharing-Modell Sinn:

Although today’s experimental vehicles are modified versions of ordinary cars, with steering wheels that eerily turn by themselves, future AVs will have no steering wheel or pedals and will come in all sorts of shapes and sizes; pods capable of carrying six or eight people may prove to be the most efficient design. Rather than work everywhere, these pods will initially operate within geographically limited and well-mapped urban areas. And they will be shared “robotaxis”, summoned when needed using a ride-hailing app. The first self-driving vehicle you ride in will be shared, not owned, for a combination of technological and economic reasons. […]

The initial deployment of self-driving vehicles as robotaxis makes sense because they are limited to a particular area—and because the sensors needed for a fully autonomous AV to sense its surroundings and figure out how to respond currently cost more than the vehicle itself. That is less of a problem for a shared robotaxi, however, which will be in use and generating revenue for several hours a day. (Private cars, by contrast, are used on average only about 5% of the time.)

​Die Kostenstruktur ist, wenig überraschend, sehr gut für Robotaxis, also AVs im On-Demand-Modell:

At the moment, travelling by Uber or another ride-hailing network costs around $2.50 a mile; but take away the driver, and that cost could fall to $0.70 a mile, reckon analysts at UBS. That is less than the $1.20 a mile it costs, on average, to run a private car (when fuel, insurance, servicing and other costs are factored in). So if robotaxis really work as advertised, many urbanites could ditch their cars and save thousands of dollars a year. UBS predicts that by 2035, 80% of people will use robotaxis in cities where they are available, and that urban car ownership will fall by 70%.

(Hervorhebung von mir)

Das sind die Szenarien, die auch in der Automobilbranche diskutiert werden.

Wie wenig man aus den eigenen Strukturen und Prozessen (und dem eigenen Ego) ausbrechen kann, sieht man schön daran, wie die deutsche Automobilbranche trotzdem weiter Diesel pusht.

Hand auf’s Herz: Wie viele „Robotaxis“ werden 2035 mit Diesel fahren?
​
The Economist weiter:

AVs would dramatically reduce the number of road deaths and, being electric, cut harmful emissions in places with clean grids. Clever routing, closer spacing between vehicles and dynamic congestion-charging could cut traffic. Like cars before them, AVs will reshape cities (a long commute is easier if you work or sleep en route) and redefine retailing (shops can come to you). Carmakers will face enormous change; instead of selling to individuals, they will supply fleet operators, or reinvent themselves as “mobility service” providers.

Deswegen versuchen BMW und co. mit ihren Carsharingdiensten Erfahrungen im Flottenmanagement aufzubauen. (Ich bin nicht sicher, wie sinnvoll das ist. Es schadet sicher nicht. Aber gegen Uber, Didi und co., die mit ihren „Robotaxis“ global agieren und auch nach Deutschland expandieren werden, hat man damit keine signifikant größere Chance. Wir werden sehen.)

Interessant auch, was selbstfahrende Autos für die Stadtplanung und Straßenregulierung bedeuten werden:

Economists and urban planners should rejoice because AVs mean that, for the first time, the unwelcome externalities associated with cars can be fully priced in. In particular, dynamic road-tolling and congestion charging, adjusting the cost per kilometre according to the time of day, level of traffic, length of trip and so on, will allow fine-tuning of entire urban-transport systems. By setting taxes and tolls accordingly, planners can subsidise rides in poor districts, for example, or encourage people to use public transport for longer trips. They can also ensure that the roads do not end up full of empty vehicles looking for riders. Such granular road-pricing is the logical conclusion of existing schemes. Some cities already have congestion-charging regimes, subsidise ride-hailing in poor areas ill-served by public transport, or impose per-ride taxes on Uber, Lyft and their kind.

​Digitale Transportplattformen bringen einen nie dagewesenen Datenschatz mit, der hervorragend für Stadtplanung und -verbesserung genutzt werden kann.

(Angezapft müssen diese Systeme sowieso werden, um Straßennutzung besteuern zu können, wenn es keinen Sprit mehr gibt, der besteuert werden kann.)

Datenschutz wird ein Riesenthema werden:

For a start, AVs will record everything that happens in and around them. When a crime is committed, the police will ask nearby cars if they saw anything.

​Selbstfahrende Autos werden de facto alles sehende Kameraungetüme auf Rädern sein. Der öffentliche Raum wird von diesen konstant aufgezeichnet oder zumindest beobachtet werden. Ein Horrorszenario für das Streetviewverpixelungsland.

Die Herausforderungen hören hier nicht auf:

Fleet operators will know a great deal about their riders. In one infamous analysis of passenger data, Uber identified one-night stands. If, as seems likely, human-driven cars are gradually banned on safety grounds, passengers could lose the freedom to go anywhere they choose. The risk that not all robotaxis will serve all destinations could open the door to segregation and discrimination. In authoritarian countries, robotaxis could restrict people’s movements. If all this sounds implausible, recall that Robert Moses notoriously designed the Southern State Parkway, linking New York City to Long Island’s beaches, with low bridges to favour access by rich whites in cars, while discriminating against poor blacks in buses. And China’s “social credit” system, which awards points based on people’s behaviour, already restricts train travel for those who step out of line.

Im Gegensatz zum gestern diskutierten YouTube-Dilemma deuten die hier angesprochenen Probleme allerdings auf vergleichsweise eindeutige politische Lösungen.