Hmm, I sorta recognise this column

And, in the longer term, urban dwellers will come to regard ownership of a car as a foolish extravagance – realising that the savings in fuel, tax, insurance, depreciation and maintenance would pay for a lot of Uber rides.

Which perhaps explains what the venture capitalists see in the company: one taxi service to rule them all. An Überservice in Nietzschean terms, in which – incidentally – Google happens to have a very large stake .

After all, Uber may one day be able to dispense with those pesky human drivers. And guess who makes driverless cars?

Well, yes. Ahem.

So, imagine what could happen if Google’s project really does bear fruit and also that Uber continues to gain market share around the world. Uber’s largest cost will be that $50k a year that they say drivers are now making part time. Scale that up to the 24/7 that Google’s cars would be able to work and that’s $200k a year and more in costs. Costs that could be stripped out by dumping the meatsacks and using the Google technology. And of course Google’s going to be standing there with a workable technology and looking for someone with the network of customers and the infrastructure to ally demand for rides with supply.

Does that make Uber worth $40 billion? All a bit speculative really but it is at least conceivable that it does.

Not the first time I’ve seen Naughton having similar ideas to those I’ve floated in an El Reg column. But then obviously this is only because we’re both judicious observers of the tech scene. No other connection at all.

109 thoughts on “Hmm, I sorta recognise this column”

  1. I don’t see driverless cars coming any time soon, mainly for reasons of cost. What it comes down to is the fact that reliability is expensive, and the levels of reliability required for driverless cars would be along the lines of those seen in the airline industry. Expensively reliable systems are cost effective for mass transport machines like aeroplanes because the same system can transport several hundred people at the same time, but to put such a system into a machine for 4 people? Something would need to change.

    I know quite a lot about reliability. In the design of an oil and gas installation there are many, many automated actions controlling the process and shutting down the plant in the event of an emergency. The devices which govern the actions (which includes the entire “loop” from initiating the action to executing it) are rated according to what is known as a Safety Integrity Level, or SIL. A SIL-1 system will fail every 100 times it is asked to do something; SIL-2 every 1,000 times; SIL-3 every 10,000 times, SIL-4 every 100,000 times. For non-critical actions you use a SIL-1 or -2, but for anything which will shut down your plant in the event of a gas leak or fire you will need a SIL-4. The reason we don’t install SIL-4 on everything is because a SIL-3 or -4 system is staggeringly expensive compared to a SIL-1 or -2. As I said, the whole “loop” needs to be rated: your gas detectors, your Process Logic Controller (PLC), your valve actuators all need to be of the same rating.

    Then you have redundancy: anything which can shut your plant down should do so on the basis of 2 out of 3 voting. So you put up 3 gas detectors (all of which are SIL-4) and the PLC will only initiate the action if no less than 2 gas detectors say the same thing. This prevents false positives, or spurious trips as they are known in the industry. I can tell you from experience that the SIL levels and redundancy levels required to achieve the reliability seen in aircraft, the nuclear industry and (snigger) the oil industry (stop laughing at the back!) is seriously expensive, and it is not simply a case of the price of electronics falling and the Chinese will take care of it all. This kit is high-end stuff, and rigorously tested, plus it needs to be maintained properly.

    I’m not saying that SIL-4 levels of reliability will never be cheap, just that we are one hell of a long way from it being so now. And I would like to see the cost of a production car with SIL-4 levels of reliability if it were made now. I’d be guessing somewhere between $500k-$1m.

  2. So Much for Subtlety

    And, in the longer term, urban dwellers will come to regard ownership of a car as a foolish extravagance – realising that the savings in fuel, tax, insurance, depreciation and maintenance would pay for a lot of Uber rides.

    I recognise the concept. Because it is what Khrushchev said to Nixon about Soviet car ownership. It was obvious that giving every Soviet citizen a car was a massive investment that the Soviet State wasn’t willing to make – few cars than Black South Africans but more tanks than the rest of the world put together. So Khrushchev came up with the idea of a highly efficient centralised rental system instead.

    Didn’t work out. But they did have more tanks than the rest of the world put together.

  3. Surely the level of reliability required should be just above the level of the average human driver; at least initially. Maybe equivalent to SIL-3?
    However, given that there will be mass hysteria from the media at every accident involving a self-driving car, I suspect that the level enforced by politicians will be considerably higher (SIL-4?) to reduce the shrieking from pressure groups and tabloids. Which will cost more money and cause more deaths by delaying the changeover.

  4. Trying to apply SIL levels to humans is a bit like trying to gauge the intelligence of a computer…doesn’t really work. The principle difference is humans exercise judgement, whereas computers make decisions. By using 2/3 voting we try to mimic judgement, but it’s still decision making. I see driverless cars as less an exercise in putting military drones on the streets, and more of one of trying to get a computer to write a novel. It might be possible, but it will be ludicrously expensive before we see an improvement over the current system.

  5. Every oil field is unique, so a new design for each. So expensive.
    Driverless cars will be mass produced, so SIL 4 only once then repeat. So cheap.
    The idea of driverless cars rented like Boris bikes seems to me an attractive offer.

  6. Every oil field is unique, so a new design for each.

    Erm, no. Just no. The kit is pretty much off the shelf. And when you consider this kit is already mass produced for the oilfield, airline industry, and nuclear industry and the fact that it is still staggeringly expensive I’m not convinced simply saying “economies of scale” is going to cut it. A lot of it is down to how the stuff is manufactured, particularly the PLCs. You’ll notice that despite the iPads and other electronic goods, we’re not flying around in Chinese airliners.

  7. TN
    It’s judgement that causes accidents. (Poor judgement, that is.)
    Replacing judgement with highly constrained rules will reduce accidents to trivial numbers.
    At a guess there will be more complaints about the car’s internal MOT system detecting a fault and leaving them at the side of the road. You’ll never eliminate false positives entirely but these will be the greatest source of complaint.

  8. Replacing judgement with highly constrained rules will reduce accidents to trivial numbers.

    Which could well end up being little more than a train on rubber wheels. Writing a novel requires judgement, not simply constrained decision making. Similarly, driving as we currently know it requires judgement. If we are to reduce the judgement, then we reduce the freedom to act. This might be good to a point, but at some stage you will not want your freedom to act curtailed to the point your “car” does little more than the monorail between the terminals at Charles de Gaulle. And my point is that getting a decision-making system to mimic critical judgement is horrendously expensive, probably more so than people realise.

  9. Very interesting, Tim (Newman). In driverless cars, which systems would need high SIL? Navigation and steering, braking, obviously. A limiter for speed would be pretty simple. I don’t know why but they seem less complicated and expensive than oil industry stuff, though I guess they work on similar principles.

    I’d imagine with not too much thought about it that the big fear with cars is a packed M1, everyone doing 80mph, and the car in front loses it. But as long as those around it don’t do the same simultaneously presumably it would be less bad than in a driver controlled scenario, where reactions etc would be slower?

    Around cities, where these things would be trialled first (I assume) you would have much lower speeds (20mph would be an advance over current London speeds) and at that speed with a belt and airbags etc you usually come to no harm anyway.

    I can certainly see it working there.

  10. Tim, I’m not sure I agree with your post about judgment and mono-railing from CDG, though. That’s precisely what I want out of a cab, and TBF it’s what I want out of all car journeys I think. As long as I retain the ability to say ‘Cancel my last, take me to X instead’ I can’t see the problem.

  11. In driverless cars, which systems would need high SIL? Navigation and steering, braking, obviously.

    Not necessarily those. Steering, braking, navigation are pretty easy to automate. The difficult stuff is the judgement: is it safe to pull out? How close to the car in front should I get? Is that a pedestrian or a lamppost? What speed should I take this corner given the wet road conditions? Can I overtake this tractor safely? When should I start braking when coming up to a junction at the bottom of a hill? Etc.

    I don’t know how you can model those at all, let alone cheaply. Perhaps if the road network was reserved only for driverless cars it would be possible, but even then you’re dealing with a lot of variables that would need to be processed extremely fast and reliably. Bear in mind that I said the entire loop must be SIL rated: so in the example of a child about to run into the road, it would need to be 3 cameras plus the PLC plus the whole brake system which was SIL-4, not just the brakes. And you’d need a separate PLC for each “loop”, combining them is a no-no. This is to ensure that a child isn’t mistaken for a litterbin and *splat*. But then you don’t want to do an emergency stop every time you pass a litterbin either. This type of visual recognition humans can do much better than computers.

  12. Yes, good points.

    I’ve just read the Guardian piece- standard sub Luddite drivel.

    I wish we could split the country in half and allow/force these cunts to live in the half where they still grow their own food and make their clothes at night by candlelight.

    Apparently it’s nirvana, so there’d be plenty of takers.

    Apart from anything else, we wouldn’t have to put up with any of their shit in the Internet.

  13. Presumably (Tim) some of that is learned behaviour rather than judgement. There is a correct distance to keep apart from the car in front, a wet road requires more care, this kind of stuff is programmable I woukd have thought. Esp bearring in mind Google maps is involved: ‘If water on the sensors, take X corner no quicker than 25mph.’

    In that regard it would be safer than many drivers.

    I accept the greater difficulty with bins/kids, but I assume some sort of motion recognition system?

    Not saying it’s easy, just that it increasingly sounds feasible.

  14. Presumably most of the safety features in the oil industry could be and once were done by people monitoring screens, valves, flows etc? Even at those huge costs, automation is better. Likewise with cars, if we could cut A&E, legal costs, insurance costs, emergency services costs, repair bills, time off work recuperating etc then maybe it would add up.

    Don’t tell Krugman, he thinks car crashes are a good thing.

  15. Esp bearring in mind Google maps is involved: ‘If water on the sensors, take X corner no quicker than 25mph.’

    3x sensors, plus the PLC…all at SIL-4. That system alone would set you back about $20k, if they were relying on feed-in data from Google Maps I doubt you could achieve SIL-4: needs to be hard wired.

    The automation works on the oilfield mainly because all the parameters (pressure, temperature, flow rate, composition, etc.) are expected within a certain range. Even then, the control room are manned and the screens watched 24/7. And operator intervention is common.

  16. “I wish we could split the country in half and allow/force these cunts to live in the half where they still grow their own food and make their clothes at night by candlelight.”

    Isn’t that the whole point of Welsh devolution and Scottish independence? Let them go off into socialist nirvanas, and let England go on a more libertarian/free trade/capitalist path? If England could get a suitably rightist government for a decade, I reckon it would drive a good deal of the yogurt knitters to the Celtic fringes.

  17. I’m still trying to work out why Proggies hate Uber, as opposed to, say, Apple or Twitter…

    Uber holds a strange position in the technology industry. “Among a legion of highly detested companies, in a highly detested business, it is the thing people hate the most,” he writes. “And the reactions against it are powerful, instinctive and often hard to articulate.” […]

    The real reason so many people hate Uber, Bobbie Johnson thinks, is that “because whatever we do, we can’t stop ourselves from making it bigger and more successful and more terrifying and more necessary. Uber makes everything so easy, which means it shows us who, and what, we really are. It shows us how, whatever objections we might say we hold, we don’t actually care very much at all.

    “We have our beliefs, our morals, our instincts. We have our dislike of douchebags, our mistrust of bad behaviour. We have all that. But in the end it turns out that if something’s 10% cheaper and 5% faster, we’ll give it all up quicker than we can order a sandwich.”

    …and I’m still none the wiser. They hate Uber because it’s convenient? Must be more to it than that.

  18. One big problem driverless cars are going to cause is change in pedestrian behaviour. Our smartphone zombies are going to just walk out in front of cars all the time because they can with impunity (the cars will screech to a halt, possibly injuring and definitely inconveniencing the occupants). Whereas now car driver/pedestrian interaction requires care on both sides, and occasionally that imperfect non-verbal communication regarding whether it’s a good idea to cross now or not.

    This also creates the perfect scenario for a new form of highway robbery. Two people can simply box the vehicle in. Do that now and one of you gets run over.

  19. Driverless cars =state control of previously private motoring. That is why it is increasingly seen as the “coming thing” as opposed to flying cars which would give more freedom to hop over govt controls. Too risky.

  20. The introduction of driverless cars reminds me of the old story about Eire joining the EU.

    In order to be good Europeans the Irish Government decided to get their cars to drive on the right, but they decided it would be better to have a trial period first.

    So for one month all cars would drive on the right and if the experiment was a success then the next month all the trucks and buses would join them.

    All those ubermotors cruising round the streets with their SIL-4 systems and along comes a drugged up chav in a clapped out old transit rushing to sign on for his JSA before he gets timed out and has to visit the food bank.

  21. And, in the longer term, urban dwellers will come to regard ownership of a car as a foolish extravagance.

    Don’t many of the childless young in London already tend to take that view?

  22. I’m flattered that TimN thinks I’m using the wisdom of Solomon when I pull out from a junction. In fact I’m making a quick calculation of speed of passing traffic and gaps between cars, supplemented by my small personal data base of experience.
    That’s basic maths, not writing war and peace.

  23. Bloke in Germany – This also creates the perfect scenario for a new form of highway robbery. Two people can simply box the vehicle in.

    That’s a brilliant point.

    It’d be like what happened when Wesley Snipes got unfrozen in “Demolition Man”.

  24. In fact I’m making a quick calculation of speed of passing traffic and gaps between cars, supplemented by my small personal data base of experience.

    You’re not calculating the speed at all. You’re estimating it at best, and making a judgement. At no point do you calculate the speed of the vehicle. And speed and distance is something you learn to judge, it is not innate: watch Arab women, who have never played sport, trying to cross a road.

  25. @Kevin B

    ‘All those ubermotors cruising round the streets with their SIL-4 systems and along comes a drugged up chav in a clapped out old transit rushing to sign on for his JSA before he gets timed out and has to visit the food bank.’

    But how does that differ from the current system, except that as things stand I only have one pair of eyes and human reactions, not 360 degree sensors and computer reactions?

    @Tim N

    ‘You’re not calculating the speed at all. You’re estimating it at best, and making a judgement.’

    Here’s where I have difficulty with this side of your argument, Tim. You may be right, but how is this ‘judgment’, so called, better than sensors and a huge onboard PC which together can calculate speed exactly and act accordingly?

  26. @Steve

    ‘That’s a brilliant point.’

    No it’s not. How will they get around the onboard laser defences?

  27. bloke (not) in spain

    To riff off of Interested, somewhat, maybe we’re looking at this from the wrong direction. Trying to get a driverless car to replace a human in a human orientated driving environment. Let’s try from the other direction & design from the strengths of an automated system. Yes, designing an autonomous vehicle to operate in an environment of autonomous vehicles would require sophisticated judgement rather than rote decision making. But that’s working to the human model, where each unit has its situational awareness limited by what it can perceive & what judgements it can make itself.
    Worth looking at our human model, now and work out what’s actually going on. How much human do you need to drive a car?
    If you think about how you actually drive, very little. Or you wouldn’t be able to listen to the radio/chat with passenger/consider whether you fed the cat, this morning/plan today’s agenda/consult internal map to choose route & adjust choice to match prevailing & thus anticipated congestion levels (including those temporary roadworks appeared yesterday)(adjust according to data coming in from traffic reports on radio) /monitor fuel, temperature, speedometer/watch for speed cameras/ check rear view mirror/ monitor behaviour of other road users/ anticipate actions of pedestrians in proximity of the highway/anticipate the unexpected/ drive car.
    Lot, isn’t it?
    Impossible to do it simultaneously.
    Not much of it’s got to do with road safety.
    So we time share & prioritise. Some things are done at a low level of attention (listen to passenger witter – only important statements require response). Some at higher. Award level of attention according to priority table (f**k me that idiot’s pulling out! Brake!)
    How much of this needs to be done at the level of an individual vehicle? If you have a system with overall situational awareness, how much can you kick up a level to the system management.* How can you assign prioritisation?

    *Example: An autonomous vehicle might need multiply redundant proximity detection. An integrated vehicle system, share proximity awareness between all vehicles in proximity.

  28. Interested,

    “I accept the greater difficulty with bins/kids, but I assume some sort of motion recognition system?

    Not saying it’s easy, just that it increasingly sounds feasible.”

    Object detection is very difficult. What we have at the moment is very crude – detection based on various point matching, but it doesn’t have context. If you run your photographs through face detection, they’ll pick up things like faces from a photograph in the background of your shot in a way that a human won’t.

    We’re very good at picking up as little but a few visual clues about an object, putting them together and getting the right answer. And that includes some context – we know that people at a bus stop aren’t waiting to cross a road, even though they exhibit similar behaviour to people wanting to cross a road.

    It’s interesting research and we’ll probably initially see inventions in driver assistance, but driverless cars in somewhere like rural Wiltshire are going to take a long time to get right.

  29. Here’s where I have difficulty with this side of your argument, Tim. You may be right, but how is this ‘judgment’, so called, better than sensors and a huge onboard PC which together can calculate speed exactly and act accordingly?

    Above a certain number variables, computers simply aren’t much good at processing huge volumes of visual data. Notice how easy it is to confuse facial recognition software by pulling a funny face, or putting a hand over one eye. And if you want computers to do this to anywhere near the level of accuracy that humans can, it’s going to be very expensive to build and prohibitively expensive to maintain.

  30. bloke (not) in spain

    And just a thought.
    TimN’s SIL4 systems are very expensive. So’s human judgement. But we load share.

  31. Interested – How will they get around the onboard laser defences?

    Duh! Mirrors, obviously.

    Or silver bell-bottomed jumpsuits, like all good future people wear.

  32. @TS

    ‘Object detection is very difficult. What we have at the moment is very crude ‘

    Yep, I don’t disagree with you or Tim, or anyone (I don’t know enough about any of this to do so), but it seems to me the operative part of this is ‘at the moment’.

  33. bloke (not) in spain

    “Object detection is very difficult. What we have at the moment is very crude ”

    Give everybody RF ID tags.
    They don’t have to contain personal data. Simply say “I’m a person. Don’t run me over”

    OK. Lots of issues. But if you want a driverless transport system you do have to think outside the box. Actually change things.

  34. Why do we want to avoid running down twerps who decide to walk into the road with their eyes on their mobiles? Eugenic, innit?

  35. Personally I think the whole thing is a massive marketing gimmick to convincen the market that Google has some value beyond that of a search engine. We don’t even have driverless passenger trains in widespread use; we don’t even have cars which drive themselves and a human on board to handle the difficult stuff (like we do with planes). Even tractors doing repetitive cultivating are not yet driverless. Yet people are talking about going straight to driverless cars from what we have now? i don’t think we’re even close. Let’s try it on the intercity trains first.

  36. Steve,

    “I’m still trying to work out why Proggies hate Uber, as opposed to, say, Apple or Twitter…”

    the proggie rulebook puts independent traders (small cafes, vinyl record shops, cab drivers) over large conglomerates. You don’t really get that with computers and social networks (even if you buy from a local PC guy, you’re still paying for Microsoft).

    They’re also just not very progressive. The proggies are actually rather conservative. They love tradition. The NYT ran a piece with a cabby who did The Knowledge, and you could just sense the awe the writer feels for the fact that someone spent 3 years on a test that is about as much use as learning how to make whalebone corsets.

  37. Whilst I agree with Tim that Google cars is a marketing gimmick, the safety aspects that he goes on about with regards cars is a bit overblown.

    The safety regulations & costs with regards to O&G and especially nuclear is waaaaaaaaay over what it really needs to be. Yes, it requires lots of fail safe equipment, but when I personally know of a device manufactured for <$100 but charged to the nuclear at $9k, even when you take into account maintenance and R&D costs and that its not a mass market item, its still overpriced.

    Safety in cars is already at the right level. See airbags or ABS brakes. They can't fail. Now car spare parts are expensive, but that's partially because manufacturers cut the price of the car to the bone and make the money on the "consumables" or spare parts. Just like ink jet printers.

    As for facial recognition, yes computers find it hard to do what we do naturally but do they need to? In many cases (other than security which is more photo matching) it doesn't need to be spot on as shown by the fact that even cheap cameras and smart phones have facial recognition built in for auto focusing. So for a car, object detection doesn't need to work out what the object is, just that there is an object there.

    Introduction of driverless cars is best done slowly. Its started already with cruise control and sensors to ensure the distance to the next car is safe. But that's taken years. It'll take years and be done with expensive cars first, but it'll trickle down over at least a decade. Just like the introduction of ABS brakes and all the other driver assist stuff. People just need to get used to it.

  38. @dearieme, because most of them won’t leave estates large enough to cover the damage done to the vehicle.

  39. SadButMadLad,

    “object detection doesn’t need to work out what the object is, just that there is an object there.”

    yes it does. I don’t expect a car to overtake a dead fox in the road. I do expect it to overtake rocks in the road.

  40. I expect a car to react the same way to a rock or a dead fox in the road, if they are the same size. In both cases it should stop and let the human decide. As knowledge and experience is built up in the software, better cars might work out if they can go round an object depending on road conditions, on-coming traffic, etc. But they will never see a dead fox and think to look out for other animals that are more interested in the fox than cars or that a rock on the road might be an indication that the area is prone to rock falls.

  41. Having recently seen what Google are showing off as state of the art at the Computing history museum in California I think Tim Newman is being quite optimistic about the success of self-driving cars. This article summarizes the problems pretty well.
    http://www.slate.com/articles/technology/technology/2014/10/google_self_driving_car_it_may_never_actually_happen.html

    The fact is that getting a vehicle from a consumer chosen point A to point B in the current built environment is not amenable to the rigid script following that a Google car can do. As a great example of the problem of just following scripts have a read of this – http://kaner.com/pdfs/ValueOfChecklists.pdf

  42. The way to approach improving driverless cars’ detection and evaluation of road hazards is to try teaching them to hunt. Use evolutionary mechanisms.

    Imagine a line of cars going by, all eyeing you up as a potential meal, if only you’ll be so foolish as to step into the road.

    They don’t actually have to hunt people, of course. But it’d be more fun if they did.

  43. When it starts working it will change the world. Car ownership, and routine self-driving really will be for the fanatics and the few people/companies it makes financial sense for. Why would you bother having something depreciating on your driveway (or where I live, on the streets originally designed for 1 car per 2 households while we now have 2 cars per household) for 95% of the time and in use the remaining 5% of the time? The current mass car ownership model is a huge waste of trillions of capital occasioned solely by the desire for transport without having to book it in advance. You could get away with a fleet of perhaps 15 to 20% of the current size (enough to meet almost all of that demand for instantaneous private transport, making ownership redundant), most of it in driverless taxi companies, that’ll take you from anywhere to anywhere, with whatever you want to take (which sometimes the car you own is not adequate for) at the touch of a button.

    A manual override will be needed, though eventually people won’t have the training to operate it. Enter the reserve army of unemployed taxi drivers, patrolling on motorbikes AA-style to get those without driving licenses out of tough spots or rescue those whose vehicle has broken down. Some kind of panic button will be needed for the highway robbery scenario. And of course if you take manual control of a roving driverless taxi that complicates the liability situation considerably. Doubtless there will be relevant insurance packages available for frequent travellers, and an option of paying a euro or two extra for everyone else.

    Roll-out is likely to start with lorries, for the reasons given by other posters – costs, and initially limitations of the technology. The cost angle increased immensely by the (often arcane) legal restrictions on lorry drivers, overnighting expenses and so on. The lorries will operate on auto control for motorway cruising, and park up in marshalling areas at each exit to be driven manually to their destination. They’ll be able to cruise indefinitely until they run out of fuel, at which point it will pull up to a semi-automated gas station where a newly-unemployed lorry driver will stick a nozzle in it, until that process is likewise fully automated. No risk of getting tachoed out in a traffic queue, late running will change from an expensive disaster to a minor inconvenience.

    As pointed out, car ownership

  44. BiG,

    The problem with that is it ignores the fact that, for whatever reason, an awful lot of people feel personally attached to their cars and treat the inside as quite private space. I think this appeal of the private car will not easily be displaced by a system of essentially public cars used occasionally by people who are unsupervised. I can only imagine the state of these machines on a Sunday morning in Manchester.

  45. I would imagine that any such vehicle designed by Google will have much the same effect on any normal person as Johnny Cab had on Arnold Schwarzenegger.

    In practical terms, Tim (Newman) is absolutely right about the safety issues. These vehicles would have to be vastly safer than humans. The reality is that no machine with the safety level of a human would ever be allowed into the marketplace.

    “The domestic robot just phoned to say it accidentally left the gas on, and we need a new house”.

  46. Here’s where I have difficulty with this side of your argument, Tim. You may be right, but how is this ‘judgment’, so called, better than sensors and a huge onboard PC which together can calculate speed exactly and act accordingly?

    Tim N knows what he is talking about, and I’m happy to back that up with my 15 years experience in PLCs and industrial automation and control (he’s already made all the arguments I would have).

    A simple answer to the above question though, is “how do you catch a ball?” Do you calculate speed and velocity then determine the number of millimetres to move your hand? No, you just catch it. Just imagine trying to build a ball catching robot. The go/no go decisions in driving are very similar.

    We’re a very long way from being able to replicate the level of judgement required for self driving cars. Making the brakes, steering, etc reliable is easy (for all practical purposes, they’re already good enough). Replacing the decision loop, however, is going to be very, very hard.

  47. Luke, I think TW was less than upset and was just cocking a snook (perhaps, I could be wrong). The ensuing discussion about the possibility I found quite interesting. Acolytes generally don’t disagree with their master. So you not only can’t say something intelligent about the issue, you don’t even know how to insult people.

  48. Tim N knows what he is talking about, and I’m happy to back that up with my 15 years experience in PLCs and industrial automation and control (he’s already made all the arguments I would have).

    Thank f*** for that! 🙂

    A simple answer to the above question though, is “how do you catch a ball?” Do you calculate speed and velocity then determine the number of millimetres to move your hand? No, you just catch it. Just imagine trying to build a ball catching robot. The go/no go decisions in driving are very similar.

    This, exactly. When you watch a cricketer run around the outfield to take a high ball, he knows exactly which direction and at what speed to run in order to reach a position where he can catch it. Or rather, he sets off and makes hundreds of micro-adjustments to his direction and speed which allows him to intercept the ball. A footballer would do a similar thing when chasing a pass, and in neither case does the player actually run towards the ball: he runs to where he thinks it will be. Hence it starts as a judgement rather than a calculated decision, and gets updated continuously.

    I thought about this when watching Arab women (and often men) crossing the road in Dubai. They appeared to have no ability to judge speed and distance, and would step out in front of cars which would have to brake hard to avoid hitting them (one theory was that they were simply arrogant and enjoyed forcing drivers to stop, but I didn’t buy this). I came away from there with the very unscientific and unproven theory that the ability to judge speed and distance is learned when young, probably by playing ball games. Small kids can’t catch or intercept a ball, but as they get older they learn how to. My guess was that because the Arabs don’t play sports when at school they struggle to judge speed, distance, and things like interception points. Just a theory.

    Anyway, this judgement is something learned and based as it is on visual data being continuously reinterpreted, it is extremely difficult for a computer to replicate without non-visual information (e.g. radar, positioning sensors) or lasers. The second YouTube video in this post (H/T David Thompson) is fascinating and would have you believe that robots can be programmed to catch and throw anything: but crucially, he explains at the beginning that the balls have sensors on and there are cameras all around relaying the information back. In other words, the system relies on cameras mounted completely separately from the vehicle (in this case, quad copters) itself. What would be truly breathtaking if the ‘copters could do this using only equipment mounted on board.

    Humans can do this already, hence we’re much better at judging things without measuring: computers need to measure and calculate, or they’re lost. And I think that’s the crucial difference.

  49. Tim Newman – you are absolutely right from the engineering point of view – but I think its Mr Ecks who hits the nail squarely on the head. This is about state control of private motoring.

    I think the whole point of so called “driverless” cars is that private individuals are not supposed to own them and all the “advantages” I;ve seen cited seem to implicitly assume this.

    An interesting variation on zil lanes and I wonder who will be allowed access to these things when the first stunt is set up in some unfortunate town.

  50. @Ltw

    “A simple answer to the above question though, is “how do you catch a ball?” Do you calculate speed and velocity then determine the number of millimetres to move your hand? No, you just catch it. Just imagine trying to build a ball catching robot. The go/no go decisions in driving are very similar.”

    I honestly bow to your greater knowledge, and Tim N’s, but this seems to me to be wrong.

    I’ve played a lot of cricket and rugby and understand about the difficulty in catching a high ball.

    However, there are many more factors involved there.

    Firstly, and perhaps trivially, if the idea was to *miss* the high ball the analogy might be better.

    In the car scenario, all we’re doing is waiting until there is no car.

    Secondly, there are various planes involved in catching a cricket ball. It rises and drops through them having started in an essentially unrestricted 360 degree journey.

    A car is moving in one direction, and is restricted (one hopes) to a drift of a few centimetres left or right.

    Thirdly, a catcher has (usually) to make ground towards where he thinks the ball will land. Outside the professional game, this is often on undulating ground, but even in the pro game the player’s head is moving, and he’s thinking about running as well as catching.

    The car is stationary at a t junction.

    Finally, a car is two tonnes of metal, and amenable to basic radar (I would assume some sort of CAS as exists in aircraft, with either that or radar as a primary system that the other as backup).

    A cricket ball is tiny and made of cork and leather. Just ‘seeing’ it is vastly harder for man or machine.

    As per Ian B and others, I don’t expect this to arrive on our streets next year. But within my lifetime I think it’s guaranteed.

  51. Although I think there are lots of reasons for it not to ever happen, I still like the idea of staggering out of a party and having my car take me home. Wait, that happens now. But I have to make it up to the wife 🙂

  52. @Tim N

    “This, exactly. When you watch a cricketer run around the outfield to take a high ball, he knows exactly which direction and at what speed to run in order to reach a position where he can catch it. Or rather, he sets off and makes hundreds of micro-adjustments to his direction and speed which allows him to intercept the ball. A footballer would do a similar thing when chasing a pass, and in neither case does the player actually run towards the ball: he runs to where he thinks it will be. Hence it starts as a judgement rather than a calculated decision, and gets updated continuously.”

    Yes, but how is that analogous to a car waiting to pull out onto a road, which is a fairly simple matter of calculating straight line speed?

    I have no idea when this thing will be feasible, but saying machines can’t catch cricket balls seems a bit like (in 1900) denying flight is possible because we can’t commute to the moon.

  53. @Interested, no, you’ve missed the point. You’re focusing on the sensors and simplifying the problem. We’re getting into AI territory here. The human brain is much better at analysing complex situations than a computer.

    What does the driverless car do when there’s a pothole? Drive over it? Drive round it? It wouldn’t have shown up on the basic radar you postulate to avoid a collision with another two tons of metal, but either could be the correct choice, depending on the circumstances.

    Robots work well in controlled conditions, and I mean very tightly controlled. Outside that… we’re just not there yet, and I have serious doubts we ever will be.

  54. @Tim,

    You’re quite right, I left Manchester over 10 years ago and have lived in largely more civilised places since. In said civilised places it works with car sharing because if you turn up to a trashed or filthy vehicle, you either complain before taking the car or the next person complains (which means you pay).

  55. @Ltw

    “no, you’ve missed the point. You’re focusing on the sensors and simplifying the problem. We’re getting into AI territory here. The human brain is much better at analysing complex situations than a computer.

    What does the driverless car do when there’s a pothole? Drive over it? Drive round it? It wouldn’t have shown up on the basic radar you postulate to avoid a collision with another two tons of metal, but either could be the correct choice, depending on the circumstances.”

    Yes, I was specifically addressing the question of cars pulling out into traffic which I thought had been raised as being analogous to catching a cricket ball?

    I may well be wrong, as was halfway through a game of Linkee with my daughters while posting that comment!

    Re potholes, the technology exists and is improving now to see depressions in rough ground – it’s used in the military in identifying IED sites – so I assume (I guess, really) it would work better with actual holes in otherwise flat surfaces?

    In which case I assume the answer to your question is, the driverless car would drive around it (if there’s another car in the opposing carriageway, it would slow and wait). Or it would do as most of us do, and straddle the pothole.

    But yes, I can see that technically it presents more of a challenge than ‘Is that car travelling at a steady 50mph going to hit me if I turn out or not?’

    However, someone about (BiG?) points out that this is maybe more of a cultural shift than a technological one.

    If we cut the cost of car ownership, insurance, fuel costs, tax, repair bills etc, perhaps the result will be more cash available to pay for road repairs.

    I personally would be happy to pay another £1,000 per annum for that if my car insurance bills vanished – I’d be quids in. I’d be quids in if I paid another £5,000, and a small fee for journeys!

    Thus there might not *be* any potholes.

    There would (or could) certainly be a very good system for reporting them, much as the traffic jam reporting system is maturing slowly into something which is actually quite useful.

    I appreciate entirely that the next question is, perhaps, what about if there’s a temporary problem – say a tree branch falls into the road?

    Well, in that case I’d expect the driverless car to stop (surely it could recognise an object in the road, even if it couldn’t name it?) and the passenger to get out and move the branch, pretty much as happens now.

    I don’t see this issue. I don’t drive over anything in the road – dead badgers, rocks (I live in a dry stone wall area), branches etc – I swerve around them or if something’s coming the other way I stop and *then* I drive around them.

    (Sometimes I do indeed get out and move stuff.)

    Tim N talked about David Thompson’s thing about drones and cameras – I can’t see any technical reason why some sort of guidance beacon system couldn’t be put in place at the sides of the roads. It would be costly, but so is everyone owning cars.

    I should add that I’m not massively hot for this idea, I just think it will come and that there would be qualified benefits if it did.

    I share Mr Ecks’ concerns re privacy, but that horse has bolted, or is bolting – before long, all cars will be fitted with black boxes, I’m sure.

    I also wouldn’t want people to be forced to ‘go driverless’, and it might well be that relatively wealthy people like myself might keep a car for personal use and use a driverless for getting back from the Dog and Duck.

    I also see Tim N’s point about vomit-filled vehicles; I assume some sort of registration system would fix that – there’d presumably be no problem in identifying the source of the vomit and charging him or her.

  56. @Me

    By the way, this:

    ‘But yes, I can see that technically it presents more of a challenge than ‘Is that car travelling at a steady 50mph going to hit me if I turn out or not?’’

    is one of the big advantages.

    People drive erratically. Too slowly if they’re old, too quickly if they’re young and trying to race their mates, or impress a girl. Some of them speed up or slow down for no discernible reason.

    They check their texts, they twiddle the radio, they turn up the heat, they see a pretty girl in a short skirt, their lane discipline is terrible, they can’t quite remember if this is the right way so they’re looking for landmarks etc etc. Their cars are often badly maintained.

    A driverless car doesn’t do any of this and ought to be perfectly maintained, given it’s the responsibility of the company you’re hiring from, and highly punishable/sueable if not maintained.

    I appreciate this isn’t foolproof, but I bet it would have better brakes and tryes than the average eighteen-year-old’s XR2 (or whatever they drive now)!

  57. Just wondering about the SiL-3 or 4 level.
    Can we not use an array of 10 level 2 sensors and just use the value of 80% of them?
    Surely that would lower the failure probability more than using 2 or 3 very expensive sensors?

    I think the biggest problems will come when driverless cars and proper cars are on the road at the same time. Drivers communicate in lots of ways, hand signals, lights, hazard lights. A lot of these are context sensitive and i can see lots of problems arising when someone gives a signal to a car which is then ignored because there is no driver.
    Also, the point made earlier about people stepping out in traffic. Can easily see this resulting in lots of hippies and hipsters getting squished…

  58. Yes, but how is that analogous to a car waiting to pull out onto a road, which is a fairly simple matter of calculating straight line speed?

    Ah, but it’s not. A car and a lorry travel at the same speed, but you know if you pull out in front of the lorry you’d better put your foot down because if something causes you to brake immediately afterwards, he’ll smash into the back of you. So you don’t do a cheeky “nip out” and then brake in front of a lorry when you’re turning into traffic.

    Also, when you’re pulling across traffic on a dual carriageway you instinctively know that a lorry won’t be speeding (they can’t), but a car might and it might be accelerating and you’re not sure by how much. A lot of this is subconscious and learned with experience, but there’s a lot more going on than simply a calculation of speed.

  59. Can we not use an array of 10 level 2 sensors and just use the value of 80% of them?
    Surely that would lower the failure probability more than using 2 or 3 very expensive sensors?

    The problem is, it’s not just the sensors, it’s the whole loop: sensors, PLC, and actuator. The whole “loop” needs to be guaranteed not to fail more than 1 every 100,000 times. Additional sensors wouldn’t help with that.

  60. @Tim N

    ‘Ah, but it’s not. A car and a lorry travel at the same speed, but you know if you pull out in front of the lorry you’d better put your foot down because if something causes you to brake immediately afterwards, he’ll smash into the back of you. So you don’t do a cheeky “nip out” and then brake in front of a lorry when you’re turning into traffic.’

    See, I think this here is a prime example of wondering whether a computer will do something a human should not be doing, cross-pollinated with a dose of ‘happens anyway’.

    If you pull out into traffic and immediately have to slam on the anchors, chances are you are driving badly.

    Yet, people do drive badly.

    I see no reason why a driverless car with rear sensors couldn’t speed up as it senses a lorry approaching – indeed, it would be able to do this while also focusing on what’s in front of it, whereas the (bad) driver who does pull out is now distracted from the road ahead, or not looking in the rear view (people only having one set of eyes).

    If the driverless car *has* to slam on the anchors because a Challenger 2 has pulled out in front of it, and then takes the hit from the lorry, well… so would a human driver.

    Additionally, sure, maybe the lorry driver is a dickhead and hasn’t seen you. But then, the *driverless* lorry wouldn’t be bombing along and would have slowed as soon as you pulled out.

    Seriously, people are the problem here, not the solution.

    ‘Also, when you’re pulling across traffic on a dual carriageway you instinctively know that a lorry won’t be speeding (they can’t), but a car might and it might be accelerating and you’re not sure by how much. A lot of this is subconscious and learned with experience, but there’s a lot more going on than simply a calculation of speed.’

    Tim, you are seriously not saying that your (or my) intuition as to a car’s increasing speed is better than a real time radar/TCAS data feed?

    I can accept some of your arguments, others I have no idea about and you’re clearly way more qualified to judge, but this one sounds most odd to me.

  61. By the way re this:

    ‘you instinctively know that a lorry won’t be speeding (they can’t), but a car might and it might be accelerating’

    Not in a driverless world, surely? The conflicting car will be adhering to the speed limit.

  62. And what I meant to say re the lorry scenario is, the days of the ‘cheeky nip out’ will be over.

    (And we’d all probably reach our destinations quicker, as erratic human driving of that sort is what causes traffic jams.)

  63. Tim, you are seriously not saying that your (or my) intuition as to a car’s increasing speed is better than a real time radar/TCAS data feed?

    No, but it is an example of where human experience can trump the data analysed by a computer: I know a lorry moving at 40mph and accelerating is not going to continue doing so beyond about 50mph. A 1992 Sierra Cosworth doing 40mph driven by a couple of chavs and accelerating is likely to be at about 80mph by the time it reaches me. A computer isn’t going to be able to judge the future behaviour of the object, whereas a human can.

  64. Here’s another one. Consider a roundabout. Using a roundabout by design involves pulling out in front of cars which are going to turn away at the last second: you can figure out by the lane, (hopefully) the indicator, the speed, the slight orientation of the car (mainly the wheels) and the driver’s head whether it will turn or carry on around the roundabout. It is very intuitive, which is why learner drivers are usually useless at roundabouts. Unless you had some system whereby cars could tell other cars what they’re going to do (good luck with getting that system operating with zero errors), I don’t know how a computer can analyse this.

    The alternative would be to do away with roundabouts and have only crossroads: but again, a lot of actions at a crossroads is intuitive (one person makes eye contact and makes the first move). You could use traffic lights, but the beauty of driving down smaller roads is you don’t need to stop and wait every time you get to an empty crossroads.

    The best I can see happening in our lifetimes is a motorway network dedicated to driverless cars which is extremely tightly controlled with limited intersections and junctions. My guess is it would be so limited you might as well lay iron rails, double the speed, and call it a train.

  65. Hmm, I sort of see what you mean. I think in aggregate though removing these decisions from human hands would be a good thing, safety wise (safety being this element of the argument against).

    My mum (and lately my dad) would *not* be able to make that calculation, and I reckon that goes for a large number of our increasingly elderly population.

    I suspect a lot of women would *never* be able to make it. If you think it’s hard to teach robots to catch cricket balls, you haven’t met my wife or daughters.

    Factor in that most accidents happen at night, and you are not identifying the Cosworth qua Cosworth, you’re thinking ‘Hmm, he looks like he’s going a bit rapid…’

    A computer can surely make that assessment much more quickly, and act much more quickly upon it.

    Of course, if a large percentage of the cars on the roads are themselves driverless then chances are the Cosworth is not going to go from 40mph to 80mph anyway.

    I’d imagine – only a guess – that more accidents are caused by people vs people than would be by people vs driverless and certainly driverless vs driverless.

    Pilot error crashes lots of aircraft. Mate of mine flies for Emirates, he says TO to landing autopilot would be waqy safer and is already doable – just that people won’t wear it.

    Appreciate this is a sample of one, and he’s not a BALPA spokesman!

  66. @Tim

    ‘ Unless you had some system whereby cars could tell other cars what they’re going to do (good luck with getting that system operating with zero errors), I don’t know how a computer can analyse this.’

    1. Who says any system would ever have zero errors? people don’t have zero errors. Just fewer errors than people would be pretty good.

    2. Technically-wise, would it really be that hard to have a system whereby car X broadcasts to car Y (and all nearby cars) that it is taking the first exit at a roundabout?

    That seems to me to be one of the more trivial problems?

  67. BiG,

    “Why would you bother having something depreciating on your driveway (or where I live, on the streets originally designed for 1 car per 2 households while we now have 2 cars per household) for 95% of the time and in use the remaining 5% of the time? The current mass car ownership model is a huge waste of trillions of capital occasioned solely by the desire for transport without having to book it in advance.”

    The problem is that the spike in everyone’s 5% usage is about the same 5% usage – 8-9 in the morning and 5-6 in the evening.

    And if you just don’t want to own a car, there’s lots of good alternatives already. For a few years, I didn’t run a car. Our family had 1 car driven by my wife. So, generally I used public transport, or took a cab. If I had to go on a long journey where I couldn’t do those, I hired a car from Avis or Budget, which is pretty cheap on a weekend.

  68. TimN says ” A computer isn’t going to be able to judge the future behaviour of the object, whereas a human can.”

    No, a computer can’t judge the future behaviour, but it can very quickly work out predicted outcomes from lots of processing. A human is unlikely to know if a Cosworth is accelerating or by how much or if it’ll be doing 80mph at a potential point of impact, but a computer will notice the acceleration over a few milliseconds, extrapolate, and adjust. The s/w will be written to be conservative so it will treat all other vehicles as lorries going at 80mph.

    Computers are crap at judgement, but fantastic at measuring lots of things very fast. Why do you think judgement trumps measurement?

    Can we stick to the point and avoid all the strawmen about throwing balls.

  69. @Stigler

    That’s two good points (about peak hours and using cabs/Avis – especially now that most hire car companies will deliver to your door).

    Is there much of a need in the personal sphere beyond wanting to get pissed and not lose one’s licence?

    I guess the main application outwith the home is any job involving driving but not much else – most bulk delivery work – or jobs where you want you/your worker to arrive refreshed rather than knackered from concentrating on the road.

  70. Computers are crap at judgement, but fantastic at measuring lots of things very fast. Why do you think judgement trumps measurement?

    You use judgement in lieu of measurement, in situations where you cannot measure every variable. If a system of driverless cars can measure every variable and accurately process the data, then no judgement is required. But by then we’d be in AI territory.

    Can we stick to the point and avoid all the strawmen about throwing balls.

    It’s only a strawman if you think it is all about measurement and not about judgement.

  71. Who says any system would ever have zero errors? people don’t have zero errors. Just fewer errors than people would be pretty good.

    2. Technically-wise, would it really be that hard to have a system whereby car X broadcasts to car Y (and all nearby cars) that it is taking the first exit at a roundabout?

    It would be hard to do it reliably, tens of millions of times without errors. And each error causing a smash. If they could get the error rate down to that of a human, all well and good. But wireless communication is not great at the best of times: for all our safety critical systems (in fact, all control systems) it is hard-wired.

  72. @Stig,

    I don’t own a car and doubt I ever will again. Get along fine with a mix of car share, taxis, and car hire, and saving around €4k a year.

    @interested,
    Even the daily commute could be handled with a far smaller pool of vehicles than we currently have. At the very least you can reduce by the number of vehicles not used for a commute at all. Further, since you point out it happens over a period of time, rather than everyone travelling at exactly the same time, each vehicle could do multiple commute-runs in that time. Even if only two, you have still halved the fleet required to meet peak demand.. Also as people will see less money going out of their hands for ride sharing, that will probably happen to a greater extent than it does now. And those that really cannot afford the risk of ever being without transport will find it worthwhile owning their own vehicle.

    Driverless cars will change the world.

  73. @BiG

    ‘Driverless cars will change the world.’

    Yep, that was pretty much the point I was making above. You have to think of this as a cultural change not a technical one. Sure, we need the tech, but that will come.

  74. Interested

    Unless you had some system whereby cars could tell other cars what they’re going to do (good luck with getting that system operating with zero errors), I don’t know how a computer can analyse this.

    2. Technically-wise, would it really be that hard to have a system whereby car X broadcasts to car Y (and all nearby cars) that it is taking the first exit at a roundabout?

    I can’t see why not?

    Ignore privacy issues for a second (and I know that’s a big ask). Driverless car has its A>B keyed in. GPS will therefore program the exact route. If Car 1 can read’s Car 2’s programmed route, it knows exactly what its intentions are on the roundabout or wherever. If it can’t (occasional wireless errors etc), it acts more prudently.

    All driverless cars could transmit (within say a couple of hundred yards) that part of their immediate local route; all other cars can act on that if they come into immediate proximity?

  75. A computer can surely make that assessment much more quickly, and act much more quickly upon it.

    Nope.

    Can we stick to the point and avoid all the strawmen about throwing balls.

    No, because they are the same problem. And it was catching balls (throwing balls is an easy and solved task).

    The best I can see happening in our lifetimes is a motorway network dedicated to driverless cars which is extremely tightly controlled with limited intersections and junctions. My guess is it would be so limited you might as well lay iron rails, double the speed, and call it a train.

    This is a very good point. It certainly won’t be the Uber model of pick me up on the side of the street and drop me there.

  76. We seem worried about the error rate and reliability of the smarts in the car. What about the error rate and reliability of the plain old mechanical stuff? So we should have SIL-4 for safety but what if a brake pipe fails? The smarts will be ok, but the car won’t. All this concentration on the tech sounds like we are getting fixated on the detail and not looking at the big picture.

    In the big picture the safety just needs to be good enough for a car to come a halt to the side of the road. It doesn’t need to be the equivalent of O&G or nuclear or aircraft where the potential for a failure to be disastrous.

  77. So we should have SIL-4 for safety but what if a brake pipe fails?

    All cars have had dual-circuit brakes since the 1970s or something: redundancy. And the pipes are tubular steel, pretty hard to break.

    In the big picture the safety just needs to be good enough for a car to come a halt to the side of the road. It doesn’t need to be the equivalent of O&G or nuclear or aircraft where the potential for a failure to be disastrous.

    Firstly, the risk will be just as large because of the frequency (as opposed to the consequences). In other words, 100 fatalities a year in separate car accidents is the same risk as 100 people killed in an annual gas explosion. Which is why car safety is taken seriously, the frequency – and therefore the risk – is very high. Fuck it up, and you’ll quickly have mass casualties.

    Secondly, designing a car that will pull safely to the side of the road is not difficult (although every road would need a hard shoulder: plenty of places where you can’t stop without blocking the road). What is difficult is getting the car to stop only when it has to and not for spurious reasons. The reason why we use 2/3 voting offshore is not to ensure the system activates when we want it to (1/3 voting would do that): it’s to avoid spurious shutdowns. This will be notoriously difficult to avoid with a car, unless people will patiently accept their car pulling over and stopping for an unspecified period on a whim for reasons the passenger cannot fathom. The purpose of the car is to transport people, not to park them on the side of the road somewhere: so you need to ensure that purpose is being met and keeping the safety stops to an absolute minimum. And that’s the challenge.

  78. Just because a brake pipe steel doesn’t mean it doesn’t fail. Unlikely yes, but it can fail. What about badly maintained tyres blowing? What about exhausts falling off into the path of the car behind? What about an alternator failure. What about a gasket leak? See loads of other issues besides the smarts.

    The risk of 100 car accidents and 1 gas explosion with a 100 fatalities is not the same. The hazard is the same.

    A car pulling over to the side of the road for spurious reasons is an inconvenience not a safety issue.

  79. Just because a brake pipe steel doesn’t mean it doesn’t fail. Unlikely yes, but it can fail.

    And that’s why cars use dual circuits, one for the back and another for the front. If one brake pipe fails, you can still brake at least two of the wheels.

    What about badly maintained tyres blowing? What about exhausts falling off into the path of the car behind? What about an alternator failure. What about a gasket leak? See loads of other issues besides the smarts.

    These are problems which have been identified and designed out (insomuch as preventing a catastrophe) years ago, as these equipment items have been present on cars for decades. Hence the car and equipment manafacturers have learned from experience, much of it bloody, how to design and maintain it.

    What driverless cars will do is introduce new systems with no track record of operation, maintainability, or reliability. Any potential problem which has not been envisaged will only become apparent when it malfunctions for the first time. And then it is back to the designers to see if they can overcome the problem to ensure it doesn’t happen again. With all the mechanical problems you cite, these have been ironed out years ago.

    The risk of 100 car accidents and 1 gas explosion with a 100 fatalities is not the same. The hazard is the same.

    That depends on the frequency of each event, and the point I was trying to make – unsuccessfully, it seems – is that the higher frequency of a road accident means the risk is “high” – as it is for a gas explosion, whose frequency is lower but consequences more severe. To illustrate:

    1) A gas explosion occurs once per year, and kills 120 people.
    2) Road accidents occur 10 times per month, killing 1 person each time.

    Both carry a risk of 120 people dead per year.

    (Yes, I do this for a living.)

  80. A car pulling over to the side of the road for spurious reasons is an inconvenience not a safety issue.

    Nobody’s saying it is. But if a car keeps pulling over to the side of the road for spurious reasons, it will be useless as a mode of transportation. So trying to overcome this problem…that’s your safety issue.

  81. bloke (not) in spain

    It’s quite odd to see TimN using the cricketer catching a ball analogy to show humans have better judgement than computers when it proves the complete opposite.
    Humans have appalling judgement when it comes to judging distances/vectors/velocities & general situational awareness.
    In the case of the cricket ball, only 1 in 10,000 would catch the ball. In a high proportion of cases the expert cricketer would be one of the 9,999. Why a magnificent catch is applauded.
    It’s why, if you’re being taught to fly an aircraft you’re repeatedly told “USE THE BLOODY INSTRUMENTS”. Because human senses are very unreliable.
    You could, however, construct a mechanism that would catch almost every ball with a high degree of reliability. There are, after all, weapons systems can hit a shell in flight. We can build systems will put an object into a small area hundreds of millions of miles away.
    The problem with a driverless car wouldn’t be other driverless cars. It would be driven cars.

  82. To be fair to Tim N, BniS, I suggested the catching the ball comparison. I wasn’t talking about athletic cricket catches either. I was more thinking of tossing something across the room to someone. The point is that people can catch unconsciously and *without* doing calculations. The brain is better processor than you think.

    Yes, pilots have to watch the instruments. But that’s just adjusting to a new set of sensors. The decision making is still made by the pilot. Flying is a much easier automation problem anyway, and largely solved.

    You could, however, construct a mechanism that would catch almost every ball with a high degree of reliability.

    Under controlled conditions, and at great expense, sure. But that’s all it would do. Toss a set of keys at it, or a box, and it will fail. It’s a much harder problem than you think.

    The problem with a driverless car wouldn’t be other driverless cars. It would be driven cars.

    Yes, the only way it would work is if you restrict the environment in which they operate to be predictable. Which is fine in a limited space (there are factories that have automated trolleys/cars transporting stuff around, picking parts, etc). I live in Australia. There are tens of thousands of kilometres of (generally shitty) roads here. That’s several orders of magnitude harder to deal with.

  83. There’s an interesting (to sad geeks like me, anyway) difference at the moment between British and US ships these days; the USN’s Arleigh Burke-class destroyers are roughly the same size and do broadly the same job as our Daring-class destroyers (leave the oversimplifications to one side for now) with one major difference being that the Darings have half the crew of the Burkes, with major impacts on spare space, crew comfort, running costs and so on (all positive in the Royal Navy’s direction).

    Why? Because the RN caught up to commercial practice of automating and de-manning routine jobs, while the USN are still welded to the notion that any gauge of any importance must be continuously monitored by a very, very bored sailor, who they insist will do a better job of noticing any deviation than an automated system. Some stuff needs people, other jobs are better done by machine.

    Having done the safety-critical nightmare myself (weapon systems, rather than oil/gas/nuclear – a really odd area of “you’re not allowed to kill anyone ever, unless you’re deliberately killing a hundred or three Russians at once” I recognise Tim Newman’s concerns, but at the same time I remember the arguments about how it wasn’t possible to introduce systems like antilock braking because “what if it fails” and “how do other drivers cope” whereas now ABS is heading for universality.

    My own take is that “driverless” will be initially limited to areas where necessary supporting infrastructure to make it work (road edge markers, roadworks with the right markers, rebroadcast units for driverless cars sharing their routes and intentions, and so on) has been installed, probably parts of London and other hotspots, and will slowly spread out from there; it’ll be either commuterville (especially if an application of “driverless” is “it arrives at your house. you drive to work hands-on or hands-off, you get out, it drives itself to its next pickup” rather than a control-less Johnnycab – which also reduces, if not eliminates, the embuggerance of ‘spurious stops’ in driverless mode) or hub-and-spoke for other transport (expect airports and railway stations to be early adopters of the infrastructure)

    The other issue with using supporting infrastructure and outside reporting (only using ‘roads recently reported good condition’ for instance) to reduce the burden on the cars is learning and derisking: if your driverless car is using roadside markers and enhanced comms, some can carry prototype guidance units denied access to those to see “are we able to operate without these aids yet?”

    None of this is new, though: Larry Niven, decades ago, had a character watching eccentrics drag-racing antique 20th century cars and wondering what the fuss was about as these quaint, smelly, noisy vehicles rattled along the restored tarmac… before being struck with horror that _those vehicles had no autopilots_ and were being controlled only by humans without radar or other input, without a benevolent computer intervening to keep them safe.

  84. Any practical driverless car that might appear within, say, the next five years would of necessity be very limited in its capabilities as, to repeat the point of Mr Ecks earlier, this is essentially a politically driven project not an engineering driven one. Its easier (politically) to change the roads and reduce the number of “obstacles” that have to be dealt with. If these things do appear I would expect to see dedicated lanes for them and whichever company operates them and rents them out would need to be guaranteed a return. Think about that for a while.

    The sort of driverless car that people imagine – one that could actually drive itself in all conceivable variants of traffic, road and weather conditions – think Will Smiths car in “I robot” – would require the integration of a lot of sensors of different types and very sophisticated AI to handle it all. I’m certainly not holding my breath.

    Such a vehicle would be a truly formidable engineering probem. This isn’t the gadget show. We’re not talking about flashy demonstrators under very controlled conditions to impress politicians and nerds.

    More to the point, such a car would also require the hardware and software to be properly qualified and certified. A true driverless car would be a whole new class of vehicle with a whole new set of safety and other concerns. The necessary standards have probably yet to be written and agreed. That in itself could take years and I suspect it could only be done after prototyes of proper driverless cars have actually been produced before the relevant parameters for safety etc can be properly defined in testable terms..

    Somebody – an engineer, not a politician – is going to have to sign this off as being fit for purpose. If that was you, what would you want to know first?

    Why do I get a horrible sinking feeling when I think about this?

  85. More to the point, such a car would also require the hardware and software to be properly qualified and certified.

    Oh no, the paperwork! *cringes*

    I suspect it could only be done after prototyes of proper driverless cars have actually been produced before the relevant parameters for safety etc can be properly defined in testable terms..

    I smell a commitee. I’ll just be over there. Way, way, way, over there.

  86. Its not a bad thing but it mustn’t be driven by politicians or rent seekers or hyped by media nerds who think they understand technology and how real world engineering development of complex systems works but actually don’t have a clue.

    There are too many examples of this costing us all a fortune and hindering proper progress. The only “driverless” cars likely to be seen in the foreseeable futre will be windmills on wheels.

    To a sufficiently detached observer this would all be very interesting. Alas that isn’t me as I am one of those milch cows who will be expected to provide the required bungs.

  87. “such a car would also require the hardware and software to be properly qualified and certified”

    Several US states, four UK towns and many other places already permit the use of driverless cars.

    Also, you are all wrong about the first class of non-prototype driverless vehicles on the roads. They won’t be wagons, or even taxis, they’ll be Amazon lockers with electric motors.

    I’m very optimistic about driverless cars I think they will be in use quite soon. All the problems mentioned are solvable. The hardware exists, the software can be written, once that is done, the costs of replicating it will be small, after the initial development tech prices always fall through the floor. Remember the first portable DVD player 10-15 years ago cost £1,000, now you can get one for about £25. Even if driverless cars are very expensive at first, sufficient taxi/delivery companies will buy them to see the prices collapse within 5 years.

    The economic advantages make it extremely attractive, making 200,000 taxi drivers redundant will alone save ~£6bn per year, plus the savings due to reduced accidents and the biggest saving, reduced congestion

  88. PS, always a good statistic to remember, there are as many licenced taxi drivers in the UK today as there were miners in 1970, about 300,000.

    Driverless cars will make the Uber battle look like nothing. in fact one of the biggest problems driverless cars will face in their introduction is from taxi drivers slashing their tyres, there’s going to be a bloodbath, but there won’t be a government to focus the blame on like in the 1980s, just the good ol’ free market.

  89. TimNewman: (Yes, I do this for a living.)

    I do it a little bit (you wouldn’t want a 1MW gas turbine blowing up), but mainly on the other side, as a user. I’m positive about driverless cars. The best way to introduce them is the same way all the other driver aids were introduced. Slowly and in the more expensive cars (or F1). This allow issues to be worked out and for the costs to be paid by those who can afford it and want it, first adopters. It will take years, possibly a decade, but it will also be a lot quicker than many think; it depends on how quickly people get used to the idea of it. So long as done slowly and in response to the public view rather than imposed and forced it will be a success.

    Tim sounds like he wouldn’t have wanted cruise control or ABS to be introduced because the risks are too high because everything he sees is coloured by his experience which is risk everywhere. Its like police thinking everyone is a criminal because all they see in their work are criminals.

    Mark: Its not a bad thing but it mustn’t be driven by politicians or rent seekers or hyped by media nerds who think they understand technology and how real world engineering development of complex systems works but actually don’t have a clue.

    You nail it.

  90. Tim sounds like he wouldn’t have wanted cruise control or ABS to be introduced because the risks are too high because everything he sees is coloured by his experience which is risk everywhere.

    Nope, I’m all for incremental improvements. But a step-change like driverless cars? Shit, we don’t even have driverless passenger trains yet. Let’s start there, shall we? Until then, driverless cars are the nuclear fusion de nos jours.

  91. F1 is not a bad idea. A driverless F1 car going round a pre-programmed track in a pre-programmed manner, needing only to avoid collision with the other cars, would presumably wipe the floor with the best drivers. And with all those sensors and calculation replacing meatsack judgment, would blow up rather less frequently.

  92. @Tim, I think a lot of train systems are effectively driverless. Certainly the ICE and TGV, the driver is basically only there to press a lever a couple of times a minute so the train knows its meatsack eyes are still alive. Not that, at that speed, you can do anything if the meatsack eyes do see something. Sure you can slam on the brakes, but the reason the average German train driver has to cope with two suicides per career is that once an ICE sees its obstacle it’s too late. Hence it’s a wildly popular way of topping yourself here.

    If the Spanish had the same kind of speed control/guidance as routinely used elsewhere the crash at Santiago de Compostela (caused by the train taking a corner too fast because the driver had forgotten to reduce speed) would not have happened.

  93. I think a lot of train systems are effectively driverless.

    I’m sure they are. So let’s remove the driver completely and run them for 5 years before we start getting all excited about driverless cars. Or are there good reasons not to do that which don’t apply to cars?

  94. Trains are a great example of a highly constrained problem space that is amenable to automation. Signals, rails, predictable performance and outcomes from actions. But try and get them to do a quick 5 minute diversion to drop someone off.

    Not the same problem, not at all.

    ABS is good, but it’s only a driver aid (much like power steering or automatic gearboxes, or instruments in a plane for that matter). We’re talking about who or what decides to press the pedal, not what happens when you press it.

  95. So Much for Subtlety

    Tim Newman – “I’m sure they are. So let’s remove the driver completely and run them for 5 years before we start getting all excited about driverless cars. Or are there good reasons not to do that which don’t apply to cars?”

    A lot of Metro systems are already run that way. It is mainly Unions that prevent goods trains following them. But five years experience? Been there:

    http://en.wikipedia.org/wiki/List_of_automated_urban_metro_subway_systems

    In a Grade of Automation 4 (GoA4) system like the Copenhagen Metro trains are capable of operating automatically at all times, including door closing, obstacle detection and emergency situations. However, on-board staff may be provided for other purposes, e.g. customer service.
    ….
    United Kingdom London Gatwick Airport -Innovia APM 100 inter-terminal people mover built by Bombardier Transportation
    United Kingdom London Stansted Airport Transit System inter-terminal people mover using Adtranz C-100 and Bombardier CX-100

    There is a long long list of places that operate trains without any people. Mainly built by Germans. But virtually everyone here must have ridden on one.

    I don’t want to get too racist over this, but if even the Brazilians are capable of doing this …..

  96. “I don’t want to get too racist over this, but if even the Brazilians are capable of doing this ….”

    You’re okay, I think that was just racist enough. 😉

    The DLR had been driverless since 1987, but as you say there are other reasons for not making the mainlines driverless. Unions are a pain in the arse and would bring the whole system to a standstill for as long as it takes to prevent the introduction of driverless trains. Also the costs of infrastructure are significant and would all be borne by one company initially, and although it looks like a lot of the development costs of driverless cars will be taken on by Google, they are a bit richer than Network Rail.

    Lastly, there’s not much need for it, the costs of driver’s wages will be saved, but that’s it. Rail is already about as safe as it can be, no passengers have died in 8 years, most deaths on the network are suicides or falling between train and platform. So very little can be saved due to accident prevention and very little from congestion prevention.

    Road transport on the other hand kills up to 2,000 people per year, 183,000 casualties and many more non-injury losses, and costs tens of billions in inefficiencies of congestion, heck, if you could get your car to do the school run without you that alone would free up countless hours.

  97. A lot of Metro systems are already run that way. It is mainly Unions that prevent goods trains following them.

    Oh yes, I know. I live on Metro Line 1 in Paris, which is driverless. I see no reason why driverless trains aren’t ubiquitous, but for whatever reason – unions, public fears, etc. – they are a rareity. What makes everyone so sure these obstacles will be overcome for driverless cars?

  98. Lastly, there’s not much need for it, the costs of driver’s wages will be saved, but that’s it.

    Plus all the disruption caused next time the unions call everyone out.

    Would be good if all the trains were replaced by driverless-capable units as part of the normal maintenance cycle. Then the next time the drivers’ unions walk out the train operators could say “Fine, stay away as long as you like! See if anyone notices”

  99. @TimN

    ‘Oh yes, I know. I live on Metro Line 1 in Paris, which is driverless. I see no reason why driverless trains aren’t ubiquitous, but for whatever reason – unions, public fears, etc. – they are a rareity. What makes everyone so sure these obstacles will be overcome for driverless cars?’

    But you’ve just identified the real issue – unions. It’s nothing to do with public fears, or else there’d be NO driverless trains. Unless we think those who take the DLR are particularly fearless!

    There’s no union to get in the way of driverless cars.

  100. Very interesting thread this.

    Regarding railways he technicalities of this are more complex than you might think and are essentially a signalling problem rather than a driving one, as the fiasco of the Jubilee line resignalling demonstrates. You also have to consider the cost of the necessary resignalling of an entire rail network to facilitate driverless trains, it aint peanuts. It can be done and perhaps eventually will be, there are incremental changes taking place which could enable it but these have themselves run into serious cost and technical problems.

    A better analogy with driverless cars in a rail context is level crossing operation. Network Rail has been attempting to introduce obstacle detection at busy crossings that are at present manually controlled and run into no end of problems. First it was too sensitive and threw a wobbly at rainstorms, bits of paper and the like and then after adjustment it seems it’s not sensitive enough, getting it right is proving much harder than thought ( although not by signallers who foresaw the difficulty ). I don’t know anything of the technicalities of driverless cars but if rail is anything to go by and it may not be, then it won’t be happening as soon as we might hope.

  101. Interesting Thornavis.

    I presume a lot of the tech from that application will be cross applicable, but it’s probably a different proposition again to do it at 60mph when every yard of kerb is in effect its own level crossing.

    My view remains that it will be doable, but when I have no idea.

Leave a Reply

Your email address will not be published. Required fields are marked *