Showing all posts tagged cars:

The Problem With Self-Driving Cars – And Self-Flying Planes

Many petrolheads are worried that self-driving cars will kill the idea itself of the car. Personally, I’m not that worried about that; horses have not been a primary means of transport for maybe a century, and yet there still seem to be plenty of enthusiasts who enjoy them. I rode my bike side-by-side with a horse just last weekend.

No, my main concern is about the process of getting to fully self-driving cars. Today, we have cars with all sorts of safety features such as radar cruise control and lane-departure warnings. These already add up to a limited self-driving capability – level 2 or perhaps even 3 in the NHTSA definition. I have used it myself on the German Autobahn, with my hire car staying in its lane, following curves in the road, and accelerating or braking with the traffic, including coming to a complete stop.

I was never able to get out a book or just climb into the back seat for a nap, though. I always had to remain alert and engaged in case of any failure of the self-driving systems. Ultimately, I found this more wearing than just driving myself, and so I disengaged all the systems, got off the clogged-up Autobahn, and found myself a twisty and more engaging alternative route where I could concentrate fully on the driving.

This is the "uncanny valley" problem of self-driving cars: in order to get to full autonomy, we have to deal with a transition period where the automated systems are not yet good enough to do the job, but they are already good enough for drivers to become distracted and disengaged from the process of driving. The risk is that when the systems fail, the driver is not able to re-establish situational awareness and take control quickly enough to avoid an incident. We saw this scenario play out with tragic consequences when a self-driving Uber car struck and killed a woman earlier this year.

This problem of operator attention is of course not unique to cars, and may in fact have played a part in the tragic Lion Air crash in Indonesia. As part of the investigation of that crash, it was uncovered that a faulty air-speed sensor may have been the proximate cause of the accident, leading to the plane’s avionics mis-identifying the situation as a stall and putting the nose of the plane down to recover the speed necessary to get out of that stall.

The FAA has now issued a directive to operators of the plane to update their manuals. The linked article includes this image, purportedly from an internal Boeing magazine, describing the steps involved in identifying and correcting this situation:

You’re a pilot, climbing on autopilot on a routine departure – so you don’t have much airspeed or altitude to play with. Suddenly for no apparent reason the plane pitches nose-down. You have to diagnose the problem correctly, disengage a bunch of systems, and recover manual control. It’s a hard task, even for trained aircraft pilots.

Now imagine the same situation in a self-driving car. You’re on the phone, thinking about something else, and suddenly the car is drifting across lane markings, straight towards a concrete highway divider.

How do you rate your chances?

It is true that these systems will only improve with real-world usage and with more data to learn from, but their marketing and usage need to be aligned to their actual capabilities and failure modes, not what we wish they would be or what they might be in ideal conditions. New interface paradigms may also be required to set operators’ experience accordingly.

Until we get there, a sensible minimum standard seems to be that we should not expect drivers to do anything as a matter of course that is difficult even for trained airline pilots.

Listen To The Heart

I am now the proud owner of my third1 Alfa Romeo! Well, technically I’ve only made the down payment: the car itself will arrive in a couple of months, so a good late Christmas / early birthday present for me.

Much as I would have loved a Giulia Quadrifoglio, Italy’s horsepower tax (spit) tips that over from "maybe if I scrimp & save" to "sorry, not going to happen". What I really would have liked would have been a Giulia Sportwagon, but Alfa Romeo in their wisdom don’t make one of those. I wonder if this is the American influence speaking, as they have made wagon versions of all their previous cars, going back to the 33 Giardinetta?

Regardless, this lack of a wagon option means that I had a difficult decision between heart (Giulia Veloce – most of the good bits of the Quadrifoglio, but with a thriftier two-litre twin-turbo four-pot, sending 280 bhp to all four wheels) and head (Stelvio Bi-Tech – basically a Giulia on stilts and with a bigger boot).

I had a good long test drive of a Giulia Super, which the dealer had ready to go. This is basically the spec I wanted for a Veloce, but in a lower state of engine tune – 200 bhp instead of 280. As with most modern cars, the Giulia features keyless-go, which I don’t have strong feelings about one way or the other. Alfa Romeo however have taken this excuse to put the start/stop button on the steering wheel, Ferrari-style, which is admittedly a gimmick, but a good one.

Straight away the impressions were good, with light and sensitive steering, reacting instantly to inputs and giving good feedback through the wheel. The Alfa is not only far lighter than the Beast, it’s also the lightest of its peers, usefully edging the BMW 3-Series, the Mercedes-Benz C-Class, and the Audi A4. You can feel this lightness as soon as you turn the wheel, with a level of agility and responsiveness the others struggle to match. The transmission sends drive to all four wheels, but in normal use it’s heavily rear-biased, which also helps.

Even in lukewarm 200 bhp spec, this motor pulls well. Again, the lack of weight helps here, but it’s a willing engine, and it sounds good too. The eight-speed ZF gearbox is technically the same unit as in the Volvo V70 I also drove, but the feeling could not be more different. Left to do its own thing, it shuffles gears quite competently, and indeed invisibly, especially in normal driving. However, should you be in the mood, there are those paddles behind the steering wheel… Slap the BMW-style gearstick left into manual mode, and you are in full control, and I do mean full. The dealer sales rep was sitting beside me, so I couldn’t get too extreme, but I tried hard to catch the system out, and I simply couldn’t.

Alfa Romeo also offers its DNA drive-mode selector, and the three modes are indeed usefully different. You start in N, for Normal, which is a good description. In this mode everything is, well, normal; the car will trickle around town quite happily without making a fuss, but will also downshift promptly if you put your foot down. D is for Dynamic mode, and everything gets just a bit sharper-edged; the engine note hardens, the transmission hangs on to gears for a bit longer, the traction control relaxes a bit (it can’t be turned fully off except on the Quadrifoglio), traction shifts entirely to the rear wheels, and so on. The test car did not have the sport suspension fitted that I specced for my own car, but that is also affected. The third option in the DNA system is A, the All-Weather mode, with the traction control primed and ready to step in at any moment. This mode can also be used for motorway cruising, as a sort of "eco" mode, where it is quiet and unobtrusive.

In addition to the shape of the gear lever, there is something else in the cabin that is eerily reminiscent of BMW, and that is the in-car entertainment system. There is a click-wheel which is almost entirely identical to an iDrive controller, driving a menu system which is also very similar to the BMW setup. Imitation is the sincerest form of flattery, and indeed BMW’s iDrive is by far the best input system I have used. I can’t get on with Mercedes-Benz’s COMAND system, and Audi’s MMI is just a mess. Tesla-style systems that hide all the controls in a touchscreen are actively dangerous. Instead, the click-wheel systems that Alfa Romeo and BMW use allow drivers to memorise common paths through the system – two clicks left, press, three clicks right, press – for repeated tasks, without taking their eyes off the wheel.

I don’t intend to spend much time with the Giulia’s onboard systems, though, since it has a very good CarPlay implementation. Upon connecting the phone, the whole screen – a very nice 8.8" TFT, merged beautifully into the curve of the dashboard – is taken over by the CarPlay display, which can be navigated both by touch, and by that wonderfully tactile click-wheel, with its direct haptic feedback.

Everything you touch in the Giulia’s cabin feels premium. In classic Alfa style, the instruments are very much focused on the driver, but all the control surfaces are pleasant, the ICE is competent but not intrusive, and all the other controls fall easily to the hand. The steering wheel in particular is a joy, adjusting easily to suit even my difficult requirements: I have long legs and so I sit quite far back, but even though I also have long arms, I like the steering wheel relatively close to me, low down but not so much that it obstructs the instruments. The seats also provide great lateral support during cornering.

When it comes to practical aspects, the boot is larger than I expected, but still compromised (as compared to a wagon) by a big cross-member that sits below the rear screen. Rear leg room is also perfectly adequate, even behind me, but not perhaps ample. These considerations persuaded me to also test-drive the Stelvio SUV.

This is basically the same car, just with longer suspension travel and a larger boot (and apparently slightly more rear leg room). This time I did get to drive the 280 bhp engine, but while I could feel the extra power, its impact was somewhat blunted by the extra weight. The Stelvio is still a light car for what it is, but it’s inevitably taller than the Giulia, and you do feel that when cornering. There is apparently an active suspension pack which mitigates some of the body roll, but it’s throwing technology at a physics problem which can be avoided by simply not jacking a Giulia up on stilts in the first place.

Ultimately what made my decision was that, after the Beast, I’ve had my fill of SUVs and want to get back low to the ground. I would have happily bought a Giulia Sportwagon, but since Alfa Romeo won’t sell me one, I think I can make the saloon work. The Stelvio is a fine car and I would have been happy with it, but the Giulia is just better at being a car.

I did make a bit of a gamble on the colour, opting for what is apparently the rarest colour in the gamut: Grigio Lipari. In the configurator it looks almost burgundy, but in person it’s a rich grey-blue, shifting between one and the other depending on the light, with reddish flecks embedded beneath the surface. I think it will look great, especially with the diamond-cut version of the classic Alfa "telephone dial" wheels, but we shall see when it arrives.

I’m just happy to have rejoined the ranks of the Alfisti.


  1. I previously owned a 156 with the last generation of the venerable Busso V6 engine, as well as a 147 Blackline with every "sport" option in Alfa’s catalogue. I thoroughly enjoyed both, and neither ever gave me any trouble. So much for Alfas’ reputation! 

How To Throw Away A Sale

When it came to be time to replace the Beast, I had a shortlist to work from. I initially wanted a Mini Clubman, in John Cooper Works trim, but was informed by She Who Must Be Obeyed that this was simply not big enough to act as the main family car.

I therefore went back to the drawing board, refined the list to two cars, and started arranging test drives.

The first dealership I visited was Volvo, as I liked the idea of a plug-in hybrid1, and nobody else seems to build them as cars – not SUVs, not city runabouts, but regular-sized family cars. The V70 seemed to fit the bill: gigantic load capacity, Volvo reputation for safety and reliability, and a plug-in hybrid drivetrain available. The Volvo setup even promised some potential for fun, as the hybrid system allows the driver to ghost silently around town on electric power alone, drive normally using the petrol engine, or combine both for a 400 bhp hit.

Unfortunately it was not possible to arrange a test-drive of the hybrid option. This was my first warning sign, especially when I tried other dealers in nearby cities and nobody had one to drive, not even fitted to a different car (the same drivetrain powers the larger V90 wagon, as well as the XC60 and XC90 SUVs and S70 and S90 saloons). I am not excited by the idea of owning some sort of unicorn and potentially having to wait for parts to be shipped from China if anything goes wrong.

Regardless, I agreed to test-drive a V70 with the popular D4 drivetrain. This is a common-or-garden turbo diesel four-pot, which sounds quite clattery from the outside of the car. On the inside, Volvo’s usual understated Scandinavian luxury muffles the racket to a distant drone, but it’s never a particularly pleasant sound.

That first impression of luxury was short-lived, though. The test-drive car had a middle-of-the-road Business spec, which looked fine but did not quite live up to its billing when it came to various important touch-points. The engine is started by twisting a little knob, which felt fiddly and flimsy. The steering wheel was nice enough, but had some very rough stitching right where the driver’s thumbs rest. The buttons all felt plasticky, too.

These may seem like quibbles, and indeed I would happily be able to put up with them on a Škoda, or indeed on something that compensated by being more driver-focused. However, the Volvo’s big selling point is that it’s a nice, safe place to be. It plays on that Scandinavian notion of hygge, which is probably best translated as "coziness", although that does not quite cover it. For a car this expensive, everything I touch has to fit into that overall impression. Instead, every touch point I came in contact with undermined that cozy feeling.

The next disappointment was the in-car stereo. New Volvos have an iPad-style portrait-oriented touchscreen high in the middle of the dashboard. This is fine, although I found it a bit distracting even on a brief test-drive. It doesn’t go quite as far as new Teslas in putting all the controls there, but it goes far enough in that direction that I would worry about not being able to access controls easily on the move without taking my eyes off the road.

CarPlay is also a must-have feature for me, and as Volvo was a launch partner of this technology, I was looking forward to testing out their implementation. Unfortunately, it took far too long to get my phone recognised by the Volvo’s systems, and once it was connected (by cable – still no wireless CarPlay 2), instead of taking over that big screen, it just created a little letterbox window towards the bottom, with the rest of the Volvo chrome still cluttering up the rest of the screen. The resulting CarPlay view was actually smaller than my iPhone X’s screen, which undermines a large part of what makes CarPlay useful in the first place.

The driving experience was pretty much as expected: very competent in a quiet way, and almost entirely insulated from the road. In typical Volvo fashion, the V70 is loaded with safety systems, all of which seem to be well thought out. I particularly liked the lane-departure warning system, which gently guides you back into your lane if you stray across a white line – unless your indicator is on, in which case it assumes you know what you’re doing. It doesn’t vibrate the wheel or do anything annoying, instead simply applying a little extra pressure to the power steering. It’s possible to override this guidance quite easily, but it’s just enough to catch your attention. Imagine Jeeves delivering a quiet cough to draw your attention to a potentially unfortunate choice, but resigning himself with a pained look if you do insist.

Despite this late break, and the truly cavernous boot ("trunk" for American readers), the test drive was a bit of a let-down. I don’t know whether my reaction might have been different if I had been able to drive the sportier hybrid R-Design spec I had in mind, but I got out of the car with that feeling I get when I hand back a rental car: "that was perfectly fine, but I’m glad I’m going back to my own car now". In other words, I had no intention of spending my own money on the thing, especially once the salesman made it clear he was extremely uninterested in offering me any reasonable trade-in on the Beast.

A pity; I was ready to embrace the green EV/hybrid lifestyle, but it seems it was not meant to be – yet, at least.


  1. Pure EVs are not really an option in Italy. I did test-drive a Tesla around the time I bought the Beast, because all of my friends and colleagues in California raved about all the state and federal subsidies they got, the access to HOV lanes, and so on and so forth. Unfortunately, not only does Italy not offer any of that – EVs are subject to all the normal taxes2 – because of the Tesla’s power rating, it also came in for an additional horsepower tax, with no discount for its non-polluting nature. In the same way, there was no equivalent of the HOV lane access, or London’s free access to the congestion charging zone. In other words, a Tesla Model S would cost in the same range as a BMW M5, and I wasn’t able to justify that even if I had had the budget. 

  2. Matters have improved slightly, although in a very Italian way: the taxes are still due, but then a partial rebate can be applied for through the regional government. It’s still not quite worth it, especially given that roll-out of the charging infrastructure around here appears to have stalled. With the recent ban on older diesels, maybe that calculation will change in time for my next car. Conspiracy theorists might also suspect that government incentives will coincidentally be introduced once FCA get around to launching a hybrid offering of their own. 

End Of Term Report

I bought a new car today. More on that later, but for now, I wanted to give a proper send-off to the Beast.

The Beast in a rare moment of looking small beside another car

Oh yes, it’s still in rude good health. I’m not selling it because of anything that is wrong with it, it’s just time to move on.

It is true that the Beast did finally blot its previously-unblemished copybook when it came to maintenance. While the Turbo engine is more reliable than the non-Turbo ones in this generation of Cayenne, it does have one notorious Achilles’ heel: Porsche engineers, in their wisdom, decided to fit two cooling pipes made of plastic that run basically right under the engine. Shockingly, this turns out to be a terrible idea, and the pipes are known to fail. Unfortunately, because it’s a horrible job – not quite engine-out, but near enough – it’s not the sort of thing you would do as preventative maintenance, or that you could tack on to routine maintenance. As an owner, you cross your fingers and hope.

Well, my luck finally ran out over the summer, and I had to get a tow back from a dinner in the wee hours of the morning, and then pay an unconscionable amount of money to have the busted plastic pipes replaced with more robust aluminium versions. I don’t think I’ve ever paid a bill that had a worse ratio of parts to labour! Then again, I got off fairly lightly compared to one owner, who had the broken pipes dump coolant right onto the tyres and ended up wrecking his car as a result.

Apart from that, though, the Beast is still a beast. Its party trick is still the mountains of torque that it can throw at any problem. It’s not that good at a standing start, because the traction control and gearbox do not get on particularly well together. I mean, it can be done, and 0-100 kph in five seconds flat is nothing to sneeze at in something that big and heavy, but the process is so unpleasant if you have any mechanical sympathy at all that it’s not really worth doing regularly.

Moving smoothly off the line is a better bet. Once the Beast is rolling, the combination of big-displacement V8 and a brace of turbochargers gives in-gear acceleration that can only be described as "violent". Motorway slip roads are the perfect environment for this sort of thing; as the on-ramp straightens out, simply floor the throttle. The Tiptronic transmission takes a moment to gather its thoughts, then drops into third, a donkey kicks you in the small of your back, and suddenly you are going very fast indeed, with the V8 bellowing and the turbochargers adding a shrill wail over the top. Around 6500 rpm – not too shabby for a big turbocharged lump like this – the Beast changes up into fourth and continues accelerating just as violently, and by the time fifth gear shows up it’s high time to back off the throttle and coast back down to the speed of the traffic.

That rush never gets old, and the looks of shock and awe on the face of other drivers when they see a big SUV move like that are just the icing on the top.

However, the Beast is getting to be an old lady, and there are more big bills lurking in its future. The timing belts will need doing soon, as will all four brake disks. I also had to send the summer tyres for recycling, and the sorts of compounds the Beast requires are fearsomely expensive, especially given the sheer acreage of rubber involved. Finally, even though it’s over ten years old and therefore benefits from the maximum discount, Italy’s horsepower tax is just too painful to deal with any longer.

I’ll miss the space, the comfort (enhanced by the air suspension), and the commanding driving position, not to mention the enormous thrust. I won’t miss the thirst, though, especially since the V8 is mostly inaudible from the cabin in normal driving, without any deep-chested burble to make you feel like you are getting your money’s worth for all that petrol you’re burning. It’s also worth noting that I used the locking diffs and low-range gearbox exactly twice in nearly four years of ownership, thereby proving every stereotype about SUVs’ typical usage patterns.

So it’s farewell Beast, and hello younger, more supple model. More on the car hunt soon, as well as the final choice!

New Frontiers In Driver Idiocy

I always get very irritated by people in new high-end cars – ones that certainly have Bluetooth, if not CarPlay or Android Auto – using their phones in their hands. What, you can’t spare thirty seconds to set up the hands-free connection once? After that you can keep your phone in your pocket/handbag/briefcase/whatever, and have both hands free for the actual driving! But no, you have to drive one-handed, or with the phone squeezed between your head and shoulder like the Hunchback of Notre Dame, squinting sideways at the road.

Today, however, I saw something that beats that. This person did have her phone connected to hands-free, and I know because everyone in the street could hear her conversation over the cranked-up stereo. However, for whatever unknowable reason, she was still holding the actual phone in her hand.

I was so flabbergasted I missed my chance to take a photograph of the madness.

Now I have real concerns about how this woman lives the rest of her life. Does she put boots on over her shoes? Or try to use knife, fork, and spoon all at the same time?

The Driver Behind The Curtain

Truly autonomous driving is an incredibly hard problem to solve. It would be hard enough in controlled situations, but in uncontrolled ones, where other road users may or may not be respecting the rules of the road1, it’s pretty close to being impossible to achieve a perfect solution. The best we can hope for is one that is better than the current state of affairs, with distracted human drivers taking an incredible toll on life.

That is the promise of self-driving cars: get the dangerous, unpredictable humans out of the loop. Getting there, however, is tough. It turns out that the tragic death of a woman in Arizona due to a failure of an Uber experiment in autonomous driving may have been caused by the uncanny valley of partial autonomy.

Let’s take it as given that fully-autonomous (Level 5) vehicles are safer than human-driven ones. However, nobody has built one yet. What we do have are vehicles that may on occasion require human occupants to take control, and to do so with very little warning. According to the crash reports, the Uber driver in the Arizona crash had no more than six seconds’ warning of an obstacle ahead, and perhaps as little as 1.3 seconds.

Contrary to some early reports, the driver was not looking at a smartphone (although more time for our phones is one of the benefits to be expected from actual self-driving cars), but at "a touchscreen that was used to monitor the self-driving car software":

"The operator is responsible for monitoring diagnostic messages that appear on an interface in the center stack of the vehicle dash and tagging events of interest for subsequent review," the [NTSB] report said.

The Uncanny Valley

I wrote about this uncanny valley problem of autonomous vehicles before:

as long as human drivers are required as backup to self-driving tech that works most of the time, we are actually worse off than if we did not have this tech at all.

In the first known fatal accident involving self-driving tech, the driver may have ignored up to seven warnings to put his hands back on the wheel. That was an extreme case, with rumours that the driver may even have been watching a film on a laptop, but in the Arizona case, the driver may have had only between four and one seconds of warning. If you’re texting or even carrying on a conversation with other occupants of the car, four seconds to context-switch back to driving and re-acquire situational awareness is not a lot. One second? Forget it.

Uber may have made that already dangerous situation worse by limiting the software’s ability to take action autonomously when it detected an emergency condition:

the automated braking that might have prevented the death of pedestrian Elaine Herzberg had been switched off "to reduce the potential for erratic vehicle behavior." Such functions were delegated to the driver, who was simultaneously responsible for preventing accidents and monitoring the system’s performance.

In other words, to prevent the vehicle suddenly jamming on the brakes in unclear situations like the one in Arizona, where "the self-driving system software classified the pedestrian as an unknown object, as a vehicle, and then as a bicycle with varying expectations of future travel path", Uber simply opted to delegate all braking to the "safety driver" – while also requiring her to "monitor the system’s performance". This situation – distracting the driver who is also expected to take immediate (and correct) action in an emergency – could hardly have been better designed to produce the outcome we saw in Arizona.

This is exactly what I predicted in my previous post on Uber:

Along the way to full Level 5 autonomy, we must pass through an “uncanny valley" of partial autonomy, which is actually more dangerous than no autonomy at all.
Adding the desperate urgency of a company whose very survival depends on the success of this research seems like a very bad idea on the surface of it. It is all too easy to imagine Uber (or any other company, but right now it’s Uber), with only a quarter or two worth of cash in the bank, deciding to rush out self-driving tech that is 1.0 at best.
It’s said that you shouldn’t buy any 1.0 product unless you are willing to tolerate significant imperfections. Would you ride in a car operated by software with significant imperfections?
Would you cross the street in front of one?

What Next?

Uber has now ceased tests of self-driving cars in Arizona, but it is continuing the work in Pittsburgh, having already been kicked out of San Francisco after one of its self-driving cars ran a red light right in front of SFMOMA.

Despite these setbacks, it is however continuing work on its other projects, such as flying taxis.

Thats seems perfectly safe, and hardly at all likely to go horribly wrong in its own turn.

Drone crash during ski race

GIF is of a drone almost crashing into a skier during a race in Madonna di Campiglio.

  1. Such as they are, yes, I am familiar with The Invention Of Jaywalking

When Robots Kill

This is not a breaking-news blog. Instead, what I try to do here is bring together different strands of thinking about an issue – hence the name: Find The Thread.

This is why I’m going to comment on the tragic story of the woman struck and killed by a “self-driving" Uber car in Arizona, even though the collision occurred more than a week ago.

A Question Of Levels

We generally talk about levels of autonomy in driverless cars. Level 0 is the sort of car most of us are used to. Particularly high-tech cars – your Mercedes S-classes, Audi A8s, many Volvos, and so on – may have Level 1 or even 2 systems: radar cruise control that will decelerate to avoid obstacles, lane-keeping technology that will steer between the white lines on a motorway, and so on. Tesla also attempts Level 3 with its Autopilot tech.

In all of these cases, the driver is required to still be present and alert, ready to take over the driving at a moment’s notice. The goal is to get to Level 4 and 5, which is where the driver can actually let go of the wheel entirely. Once Level 5 is commonplace, we will start seeing cars built without manual controls, as they will no longer be required.

The problem, as Benedict Evans points out, is that this will not be a universal roll-out. As I have written myself, autonomous driving technology is likely to be rolled out gradually, with easy use cases such as highway driving coming first.

This is the nut of the issue, though: as long as human drivers are required as backup to self-driving tech that works most of the time, we are actually worse off than if we did not have this tech at all.

In the first known fatal accident involving self-driving tech, the driver may have ignored up to seven warnings to put his hands back on the wheel. That was an extreme case, with rumours that the driver may even have been watching a film on a laptop, but in the Arizona case, the driver may have had only between four and one seconds of warning. If you’re texting or even carrying on a conversation with other occupants of the car, four seconds to context-switch back to driving and re-acquire situational awareness is not a lot. One second? Forget it.

In tech circles, self-driving tech is mostly analysed as a technology problem. Can we do this with cameras and smarter processing, do we need expensive Lidar rigs, who has the smartest approach, and so on. This is all cutting-edge stuff, to be sure, and well worth investigating anyway. You can then start speculating about the consequences if this tech all works out, and I’ve had a go at thinking about what truly self-driving cars may imply myself.

Beyond The Software

There is a whole other level beyond the technological one, which is the real-world frameworks in which these technologies would have to operate. The sorts of driving licenses we issue to humans already focus more on the rules of the road than the techniques of driving. You can learn the mechanics of driving in a few hours, especially with an automatic gearbox. The reason we don’t give people licenses after a day of instruction is that we also require them to understand how to drive on public roads shared with others.

This tragic accident in Arizona has shifted the conversation to whether it is possible to sue an autonomous car. I am working with some major automotive manufacturers, and all are developing self-driving tech – but none are prepared to roll it out, or even discuss it much in public, until these aspects have been sorted out. Car-makers are a fairly conservative bunch, used to strict product liability laws.

In contrast, the software industry by and large accepts the idea that a click-through waiver absolves you of all responsibility for your products. That is not at all how the automobile industry operates. Even strictly software faults are held to a level of scrutiny unknown in the general software industry, outside of specialised applications. In the case of Toyota’s unintended acceleration problems, the car-maker was ultimately held responsible in court for a fatal accident, due to identified bugs in its electronic throttle control system – and to the fact that code metrics indicated the probability that other, as-yet unidentified bugs were still present in the codebase for that system.

Jamie Zawinski has some typically acerbic commentary:

Note that the article's headline referred to the woman killed by the robot as a "pedestrian" instead of a person. "Pedestrian" is a propaganda term invented by the auto industry to re-frame the debate: to get you to preemptively agree that roads, and by extension cities, are for cars, and any non-car-based use is “other", is some kind of special-case interloper. See The Invention of Jaywalking.

Semantics aside, I have one question that I think is pretty important here, and that is, who is getting charged with vehicular homicide? Even if they are ultimately ruled to be not at fault, what name goes on the court docket? Is it:

  • The Uber employee - or "non-employee independent contractor" - in the passenger seat?

  • Their shift lead?

  • Travis Kalanick?

  • The author(s) of the (proprietary, un-auditable) software?

  • The "corporate person" known as Uber?

Good question, and one that so far remains unanswered.

Why The Rush To Autonomous Cars?

Finally, let’s remember that there are two reasons that the industry is storming ahead with self-driving tech. The public reason is the presumption of increased road safety through the removal of distracted human drivers from the road. However, as the complexities involved in moving beyond simple demos in an empty parking lot become clear, people are starting to suggest ridiculous solutions like "bicycle-to-vehicle" communications – in other words, instrumenting cyclists so that they will advertise their position to cars. And if you give sensors to cyclists, why not pedestrians too?

This is a typical technology-first fix: if you can’t solve the problem one way, by detecting cyclists through sensors, you solve it another way, by fitting sensors to the cyclists themselves. Here again, though, we are not in a purely technological domain. This blinkered view is why self-driving cars won’t save cyclists, at least until the thinking shifts around the whole issue of cars in general.

Here is where we come to the second reason behind the urgency in the development of self-driving tech: Uber’s business model depends on it. Right now they are haemorrhaging money – over a billion-with-a-B per quarter in 2017 – in a race to achieve market dominance before they run out of cash (or investors willing to give them more). Much of that cost goes to their human drivers; if those could be replaced with automated systems, the cost would go away at a stroke, and they would also achieve much higher utilisation rates on their fleet of vehicles.

In this view, self-driving cars are both an offensive move against Uber’s competitors, and a defensive one in case the likes of Google get there first and undercut Uber with their little pod-cars.

This sort of thing is catnip for futurists and other professional speculators, existing at the nexus of technology and business model that is Silicon Valley distilled to its purest essence. However, as the real-world problems with this project become more and more visible, people are starting to question whether self-driving cars are actually a distraction for Uber.

The bottom line is that right now we are pushing forwards with self-driving tech in the hope it will make our roads safer. This is a valid and important goal, to be sure – but those claims of increased safety from self-driving tech are still assumptions, very much unproven, as the tragic death in Arizona reminds us.

Along the way to full Level 5 autonomy, we must pass through an “uncanny valley" of partial autonomy, which is actually more dangerous than no autonomy at all.

Adding the desperate urgency of a company whose very survival depends on the success of this research seems like a very bad idea on the surface of it. It is all too easy to imagine Uber (or any other company, but right now it’s Uber), with only a quarter or two worth of cash in the bank, deciding to rush out self-driving tech that is 1.0 at best.

It’s said that you shouldn’t buy any 1.0 product unless you are willing to tolerate significant imperfections. Would you ride in a car operated by software with significant imperfections?

Would you cross the street in front of one?

And shouldn’t you have the choice to make that call? This is why, despite claims that the EU’s strategy on AI is a failure, I like their go-slow approach. Sure, roll out 1.0 animoji or cat-ear filters, but before we rely on computer vision not to run people over, or fine them for jaywalking or whatever, we should maybe stop and think about that for a moment.

Surprise: Uber, Lyft Actually Increase Traffic

A big part of the promise of the "ride-sharing" companies was that they would reduce congestion by increasing utilisation of cars. Of course, this idea of vehicles in near-constant use for a sequence of rides was hardly a great innovation, being previously known as a taxi, but there you are. However, now it turns out that Uber et al might actually be increasing congestion:

One promise of ride-hailing companies like Uber and Lyft was fewer cars clogging city streets. But studies suggest the opposite: that ride-hailing companies are pulling riders off buses, subways, bicycles and their own feet and putting them in cars instead.

I am glad that at least the description for these services has switched from ride-sharing to ride-hailing. As I have written before, Uber is nothing to do with sharing:

It is of a part with other examples of "sharing economy" (scare quotes very much intended). Very little of Airbnb is people renting out their spare rooms; much of it is people renting out stables of properties, purchased for the explicit purpose of renting them out through Airbnb. There are even companies like Airsorted that will take care of the whole process, letting landlords sit back and take in the proceeds - while conveniently forgetting to pay taxes on their earnings.
This is not "sharing" economy; these are fully professionalised marketplaces. The real sharing economy equivalents are BlaBlaCar or Couchsurfing. These offerings really do enable non-professionals to share something.

The article is fairly reasonable and balanced, if hardly ground-breaking. CityLab has been covering Uber’s unsustainable impact on traffic for a while now.

What I found really disingenuous was the last paragraph of the AP piece:

At least one study did not pin increased congestion on hailing services. Seattle-based firm Inrix scoured data from 2012 to 2015 in London and found the number of passenger vehicles, including Uber cars, remained the same or even dipped slightly. Reasons for increased congestion included a surge in road construction and delivery trucks dropping off online purchases.

This misses the point entirely. The whole premise according to which Uber et al were going to reduce congestion was that, by increasing utilisation (trips per vehicle), they would reduce the number of cars needed for those trips, and therefore the need for parking spaces and so on and so forth. If the number of vehicles remains the same, but utilisation is way up (Uber drivers gotta eat!), congestion will certainly increase!

It’s the same effect as those online purchases being delivered: instead of being chunked into a small number of large deliveries to shops, all those packages are being dropped off one by one at people’s homes, and probably by different delivery companies too. It’s more convenient for shoppers, sure – but it is not more efficient.

Anyway, I have to wrap this up and go get on the Metro, which is still the best way to get around.

See you underground!


Image by Hanny Naibaho via Unsplash

Goddess

Citroën DS 19, at Museo dell’Automobile in Turin.

The Déesse has long been one of my dream cars, but I can’t do it justice in words. Fortunately, Roland Barthes did it for me:

La nouvelle Citroën tombe manifestement du ciel dans la mesure où elle se présente d’abord comme un objet superlatif. […] La «Déesse» a tous les caractères (du moins le public commence-t-il par les lui prêter unanimement) d'un de ces objets descendus d’un autre univers, qui ont alimenté la néomanie du XVIIIe siècle et celle de notre science-fiction: la Déesse est d'abord un nouveau Nautilus.

I don’t know whether Renaud Marion had this maquette in mind when he came up with his Air Drive series, but it seems probable. Unfortunately, "Air Drive" did not include a DS, but Jacob Munkhammar has stepped up to fill that gap.

There’s a dealership I’d love to visit! I wonder whether that’s the one where they got the flying DS taxi for Back To The Future II?

Category Error

In discussing my last post about self-driving car technology, it was pointed out to me that I was being unduly pessimistic about the prospect of really smart people solving the problem of dealing with the dynamics nature of city traffic. Clarke’s First Law comes to mind:

When a distinguished but elderly scientist states that something is possible, he is almost certainly right. When he states that something is impossible, he is very probably wrong.

I am neither distinguished, nor (yet) elderly, and nobody’s idea of a scientist, and I certainly do not mean to imply that self-driving car technology is impossible. What I am saying is that adoption of self-driving tech, and indeed of any new technological offering, is not just a technology problem, and perhaps not even primarily a technology problem. For this reason, even more than because of the technical difficulties involved in navigating busy and unpredictable city streets, I expect the initial uptake of self-driving cars to be in more controlled environments, whether dedicated roadways, or inside industrial plants, airports, and the like, where random unexpected foot or bicycle traffic is not a factor. Actual go-anywhere self-driving cars will still not show up on public streets even some time after the technology has been proved out in those environments.

That said, there are signs of movement. The US has chosen ten proving grounds where this technology can be tested in real-world conditions. Also, a possible contradiction to my expectation of initial uptake in controlled environments is unpredictable Indian road traffic - and yet, Tata have announced their intention to test in Bangalore.

Just in case I was overly pessimistic, therefore, I wanted to run a quick thought experiment taking the opposite position: assuming that true, go-anywhere self-driving functionality becomes available - what then?

Everything changes

The main factor to take into account is that everything will change. To think of this future state in terms of "driverless cars" is to miss the point by as much as those who described early automobiles as "horseless carriages". It is natural to think of radical new products in the context of existing categories, but truly significant innovations define their own categories, with wide-ranging consequences.

If personal flying cars will ever be possible, it is a foregone conclusion that the results will not be what was depicted on the cover of 50s pulps, with the nuclear family setting off in Dad's finned and chromed sportster for a day of wholesome fun. The sheer traffic control and public safety requirements of a world where anybody can afford a flying car would act to limit uptake and adoption. In fact, pretty much the only way that personal air vehicles could ever become widespread would be if they were autonomous pods, with no fallible humans in the control loop. Either a central traffic control, or peer-to-peer connections between the vehicles (or a combination of the two) would be required to make that situation even remotely practical. At that point, would anyone bother to own one of these air transit pods, or would we just summon one as needed?

Many of the same factors apply to autonomous road cars. Right now, privately owned cars are idle for most of their lives. They are parked overnight, then driven to work, parked there all day, and driven back home to be parked. If the car becomes autonomous, it does not need to park near its owner; it could drive itself to a distant parking structure, or even right back home, to wait until it is needed again. At that point, why not enable other people to share the use of the car while its owner is otherwise engaged? And the logical consequence of that situation is, why own the car in the first place? Just summon it to where you are, and dismiss it when you’re done.

Right away, there is no longer any need for parking lots near offices and retail centres, just much smaller pick-up and drop-off areas. What does this do to the fabric of cities? Imagine every parking lot replaced by a park - or a reduction in sprawl, as new, much denser residential and commercial development can take the place of redundant parking structures. Even where new greenfield builds are required, they can become more efficient, no longer requiring as much space for all of those parking spaces.

Share - but how?

Some of the analysis also assumes an increase in ride-sharing. In this scenario, each vehicle has multiple occupants, which reduces the overall number of vehicles on the road, in turn reducing the need for road infrastructure. I’m not sure that this is plausible. Rather than this sort of simultaneous ride-sharing, with many people in the car at the same time, I think that the plausible future involves sequential ride-sharing, where the vehicle is in near-constant use (at least at peak times), but only by one person at a time.

The move from a large number of inefficient, single-user assets to a much smaller number of highly efficient assets with high utilisation rates will also have a dramatic impact on the automotive industry. The future of autonomous cars is not a Tesla in every driveway. It looks much more like Paris’ new driverless minibuses - in other words, the use case that we should extrapolate from is public transport, not private cars. The advent of autonomous self-driving technology will not add new capabilities to private cars. Rather, it will lead to an increase in the flexibility and capability of public transport networks.

This change will determine the end of the automotive industry as we know it. Already today, nobody cares especially about the marque of their Uber or Lyft vehicle1. In a scenario where those are the only types of vehicles on the road, the basis of competition between manufacturers would change dramatically to become something much more similar to the commercial vehicle market. The relevant drivers for competition would become the cost of operation and maintenance, without any particular brand cachet or driver experience factoring into selection. The overall size of the market would also shrink dramatically, as increased utilisation rates for individual vehicles lead to a requirement for a much smaller number of vehicles overall. Whatever happens, the sales volume for the industry will crash and most of the current manufacturers will exit the market one way or another.

Another expected advantage of autonomous vehicles is the new traffic control capabilities that they enable. One of the most frustrating types of traffic jam is the one that there is no apparent reason for. Traffic slows and stops, restarts, inches along - and then suddenly it’s flowing again. What happened in that situation is a signalling cascade: one human driver hits their brakes, the one behind them, unable to gauge intentions or speed accurately, hits their brakes a little harder, and pretty soon the whole line of cars has ground to a halt.

Instead, autonomous vehicles could enable swarming behaviours, where a whole line of vehicles can tailgate each other, safe in the knowledge that none of them is going to do anything unexpected. The immediate benefits would include optimal utilisation of road surfaces and reduced fuel consumption (from aerodynamic and other effects).

…but not today

All of this describes the ultimate end state, but for the reasons I discussed, I do not expect all of these consequences to manifest in the short term. What I do expect is that limited application of autonomous driving technology will deliver some initial benefits, and over time the technical, legal, and social hurdles will begin to fall, enabling some of these second-order benefits. However, none of that will happen in three to five years or even ten years, as some of the boosters argue, regardless of the technical progress that is made.

It is still important to think about the impact of technology adoption. Too often, IT people especially focus only on technical feasibility, and assume that just because something is possible, it will then be adopted. In actual fact, history is littered with the bones of products that failed to gain traction because of non-technical factors. Meanwhile, tech industry commentators lament the success of “inferior" products (whether or not they actually are inferior) that focus on user needs.

The canonical example of "not getting it" is of course Slashdot founder Rob Malda’s reaction to the iPod launch:

No wireless. Less space than a Nomad. Lame.

Of course we know how that played out: regardless of the technical specs, the iPod met users’ needs so completely that it defined the entire category - before, of course, it was in its own turn subsumed by the Next Big Thing.

The same will happen with driverless cars. Solving the tech issues is only a part of the problem. To achieve the sort of transformative effects that I described above will require a concerted push into all sorts of areas that I see as being currently ignored: city architecture, transportation policy, legal issues, insurance, and so on and so forth. The technology industry has a tendency to dismiss these types of issues as "soft factors".

When launching a new technology, ignore the soft factors at your peril.


Images by Austin Scherbarth and Peter Hershey via Unsplash


  1. Except that if I pay for Uber Black, I would be very disappointed it an UberX-grade Prius or whatever were to show up, instead of a big German luxury car.