back to article Fog off! No more misty eyes for self-driving cars, declare MIT boffins

MIT brainiacs have come up with some new fangled technology that could help self-driving cars cope with misty mornings. Adverse weather conditions such as rain or fog has long bedeviled autonomous vehicles, with studies showing that today's robo-rides will, at best, pass control back to their human masters when visibility …

  1. John Smith 19 Gold badge
    Unhappy

    Obvious question. Do Californian cars have rain and fog sensors so they know it's happening

    Or do they need the LIDAR output to go on the blink before they realize this?

    I got a shedload of downvotes for commenting that Uber seem to be the only company whose driverless car experiments have actually killed someone.

    By that I meant someone who had nothing to do with testing. I view anyone sitting inside a vehicle whose motion is controlled by this technology chose to take part. And have no illusions this tech is still experimental.

    OTOH Every other road user (unless the vehicles path is clearly marked in advance) is just a potential unwitting victim. They have no choice.

    1. Whitter
      Boffin

      Re: annother obvious question

      is that statistically correcting for distortions will likely require guessing the shape/geometry in question (e.g. is it a flat circular object (a sign) or a sphere (a ball)). What if it gets the guess wrong?

  2. Anonymous Coward
    Anonymous Coward

    What's the Point?

    I think the tragic, fatal incident involving Uber has just rendered a lot of developments like this pointless. It's entirely possible that we're not going to be allowed to have cars at autonomous level 4, 3, and there's doubt about some aspects of 2. In other words, it's 5 or (effectively) nothing.

    LIDAR that's able to peer a bit better through fog is a good thing (nice piece of work), but to be honest that is a fairly trivial problem in comparison to the biggest challenge of all. To get to level 5, i.e. guaranteed totally reliable full time autonomy (for some value of 'guaranteed'), requires a step change in the design approach, system specification and testing, and the cost of implementation. It becomes a full safety critical system development.

    As a species we're not so good at those. We don't even do them for airliners (which is why there's supposed to be two wide awake, sober, competent and attentive pilots at the controls able to take over when things go slightly wrong, and even then they sometimes screw up - AF447). Doing one to solve the "driving" problem is going to be phenomenally difficult. We can't even write down what "driving" involves, never mind come up with a comprehensive and testable requirements specification.

    We can kinda do it for trains, but then trains operate on closed, deserted tracks with little concern for the surrounding environment. For cars on the roads where there's horses, animals, potholes, kids, cyclists and pedestrians, forget it.

    We'll see what happens over the next few years. As the rate of testing goes up there's the grim possibility that more deaths will occur. That would focus legislators minds on whether or not the human race could be trusted with a level 4 or level 3 car; I suspect they'd conclude we cannot.

    1. LucreLout

      Re: What's the Point?

      I think the tragic, fatal incident involving Uber has just rendered a lot of developments like this pointless. It's entirely possible that we're not going to be allowed to have cars at autonomous level 4, 3, and there's doubt about some aspects of 2. In other words, it's 5 or (effectively) nothing.

      I watched the footage and must disagree.

      The car obviously gets it wrong and should have executed an emergency stop. However, my main takeaway from it was the in car footage of the meatsack. Said meatsack was quite obviously spending 90% of their time faffing with the car and considerably less than 10% watching the road. How then, can they be expected to competently take over and resolve a situation if for near all the time they have no idea anything is happening?

      If someone needs to pay attention to in car screens, then they should be removed from the car and data relayed to a lab, leaving the 'driver' to focus on the mission critical part of being a driver - safety. Everything else is secondary to that.

      This whole accident could and should have been avoided, in my view, if the person responsible for the vehicle was paying sufficient attention to the road. The golden rule is simple in explanation and use: You must be able to stop in the distance you can see to be clear and reasonably expect to remain so; half the distance on a single track road.

      In the interests of full disclosure, I have zero experience of working on self driving vehicles and little knowledge of how the software works; I am an advanced driver (beyond a basic IAM test), with moderate levels of track based racing experience.

      1. ThatOne Silver badge
        FAIL

        Re: What's the Point?

        > This whole accident could and should have been avoided, in my view, if the person responsible for the vehicle was paying sufficient attention to the road.

        That's a totally unrealistic expectation IMHO. Drivers often have problems staying focused even in classic manual vehicles, so it's impossible to expect them to do that when a car is driving itself and the "driver" has nothing left to do.

        Either the car can drive itself reliably and the humans inside can do whatever they want, or the person supposed to be responsible for the car has to keep driving it to some extent. Everything else is just nonsense: Someone who has been doing nothing car-related for long boring hours will not be able to hit the brakes in a split second if need arises. Whoever pretends otherwise is either terminally naive, or looking for a convenient scapegoat.

        Luddite? No, realist.

        1. Anonymous Coward
          Anonymous Coward

          Re: What's the Point?

          Precisely. I've long believed that one of the benefits of driving a manual transmission car is that it is more engaging - pun intended. It pulls my mind closer to the process on more than one level. I.e., it's more fun. Bored drivers are distracted drivers, because the truth is that once learned, driving is an easy task to master - superficially - and to conduct with minimal attention during the overwhelming majority of time spent doing it. To expect "safety" drivers to regularly and attentively monitor autonomous vehicles, while such vehicles continuously and counter-intuitively signal that drivers are just baggage is unrealistic.

          If I've got to watch the road, make it entertaining.

        2. LucreLout

          Re: What's the Point?

          That's a totally unrealistic expectation IMHO. Drivers often have problems staying focused even in classic manual vehicles, so it's impossible to expect them to do that when a car is driving itself and the "driver" has nothing left to do.

          This isn't a product in commercial use - its a product being developed and tested in a public space. The meatsacks primary responsibility is everyone elses safety and the development and testing of their product comes only after they've done so.

    2. Anonymous Coward
      Anonymous Coward

      Re: What's the Point?

      "I think the tragic, fatal incident involving Uber has just rendered a lot of developments like this pointless."

      And how many people were killed yesterday in accidents where cars were under the control of a human driver? How many of those would have been prevented if the car were instead driven (or even supplemented) by autonomous technology (even today's state of the art). Yes, the video is disturbing, but we can find plenty of disturbing video from poor human drivers as well.

      As the rate of testing goes up, I guarantee the death rate involving autonomous cars will go up, it's basic math. At the same time, the death rate involving human controlled cars will begin to drop. In the long run, getting us unreliable, distracted, sleepy, cellphone watching, radio listening, attention wandering humans off the task will save lives.

      I agree, as a species we are not good at designing full time safety critical systems. Unfortunately for us, we're even worse at actually being the full-time safety critical systems.

      Keep this number in mind: ~40,000 annual deaths. That's a ballpark figure for how many traffic deaths we have in the US. The goal needs to be for autonomous driving to result in killing fewer than 40,000 people per year, not some esoteric trolley problem discussion.

      AF447 is a terrible example to argue for humans over automated control: the plane knew what to do, but the meatsacks overrode and ignored the computer. From the old joke about automating airplane controls, Airbus simply forgot to include the dog (the dog, of course, being there to bite the pilot if he tries to touch the controls).

      1. Anonymous Coward
        Anonymous Coward

        Re: What's the Point?

        AF447 plane "knew" what to do? How do you figure? It "decided" to disconnect the autopilot in the face of conflicting sensor data. Had it not been for the automation, perhaps the crew would have been attentive enough to properly diagnose the pitot problem earlier, as crews are supposed to do. Rather, they were sucked into the problem, given a puzzle to solve while disoriented. Air crews are supposed to be prepared for this, ready to override autopilots, and in this case they failed. Imagine the more casually trained and less skilled operator of barely autonomous road vehicles in similar situations with less time to react.

        The computer needs to be much better and far more capable of autonomy for this to become a practical and safe alternative to full human control. There is a chasm to cross.

    3. Francis Boyle Silver badge

      Why would level 4 be ruled out?

      If it turns out that AVs can't cope with pedestrians then restricting them to areas where pedestrians don't have access (which is a subset of level 4) is the obvious solution.

  3. Steve Davies 3 Silver badge

    What about the Winter sun

    that is very low in the sky just when you want to drive home

    Try going south along the A3 past Guildford when the January sun is setting. As you get close to the top of the hill, the sun is right in your eyes. If they can manage that test then I'll accept that they are ok.

    I'm sure that other posters have their own bit of road that they love to hate at certain times of the year.

    1. Pen-y-gors

      Re: What about the Winter sun

      Very true, and I suspect technology will be better at coping with low sun (filters, lidar etc) than a poor blinded human.

      1. Geronimo!

        Re: What about the Winter sun

        Driving a new Opel (Vauxhall) Insignia with all those nice and helpful toys onboard, I can confirm: Being blinded by the sun is no fun of you're the driver. My ACC (Adaptive Cruise Control) does recognise the car in front of me w/o any issue an slows down to match the speed.

        Or hits the brake and warns you - that happend once as well. Fun, if the car in front of you has no working brake lights and you are blinded by the sun...

        1. Craigie

          Re: What about the Winter sun

          ACC in my Ioniq has so far only switched off in heavy snow fall.

    2. Anonymous Coward
      Anonymous Coward

      Re: What about the Winter sun

      "I'm sure that other posters have their own bit of road that they love to hate at certain times of the year."

      M42 J2 to J3

      Easy to tell when it's low sun, the speed on a clear day drops to about 5mph.

      Apparently sunglasses are only to be worn in summer on top of your head, or in nightclubs.

    3. Jason Bloomberg Silver badge

      Re: What about the Winter sun

      I was driving home one summer's day with the sun behind me when I was completely blinded by the sun reflecting off the back windscreen of the car in front. I couldn't see anything and, as approaching a roundabout with a car in front, all I could do is a near emergency stop. From the blaring horns I doubt the cars behind had any clue why I had 'slammed my brakes on' or why I was now 'parked' yards from a roundabout while my retinas recovered.

      As to the Uber fatality - While the police say there appears to be no fault on the part of Uber I cannot agree and think that call is premature. I believe an attentive driver would have seen the victim and at least have braked hard before the collision. The Uber doesn't seem to have done anything and supposedly has eyes which can see in the dark better than any human.

      As to the attentiveness of the 'driver' in an autonomous Uber - While they should be ready to act instantly; I think we all have to accept that it's not realistic to expect that at all times. They certainly should not be used as defacto scapegoats. The demonisation and character assassination from the Daily Mail and elsewhere I find quite disgusting.

      1. tiggity Silver badge

        Re: What about the Winter sun

        I have not read what the Fail etc. have had to say about this, I can guess the tone, but sadly it's a dream story for gutter press to salivate over.

        "Robot" car kills person

        Dead person appears to have substance, homlessness etc. issues

        The "backup human" turns out be trans (identifying female) but has previous armed rbbery conviction when identified as male.

        Driver is not "white"

        Getting the redtops etc. to ignore that is like trying to stop a (male) dog licking its balls.

  4. Anonymous Coward
    Anonymous Coward

    Fog, rain, the wind blowing leaves, snow, a few small rocks falling from an overpass. The last would be exceptionally rare and probably best just to drive through rather swerve or brake and hit the car beside you or get rear ended. What happens when the system has been working appropriately for a couple of years and then fails and the eighteen year old has to take over and actually drive without ever having had to drive before? I think we are a long way from systems reliable enough and cheap enough to be used outdoors amongst the public.

    What about systems that are used more like we meat sacks do? We sit inside and use our binocular vision, so what about smal binocular cameras mounted on each side of the wind shield at the top corners and the same for the back? Lidar sensors could be mounted there too. That would keep precipitation from building up on the sensors.

  5. Anonymous Coward
    Anonymous Coward

    What did they expect? In a self driving car the person behind the wheel is no longer a driver but a passenger with an attention span to match. From the footage the driver was either sleeping or staring at a mobile. As for the victim females on a bike never look to oncoming traffic and a car is relatively small they tend to miss complete trams with deadly consequences.

  6. Paddy

    Red flag laws ...

    In the 19 century the UK had laws ( http://mentalfloss.com/article/71555/ridiculous-uk-traffic-laws-yore), restricting the first motorcars: 2 miles and hour in town; Have a man walking in front waving a flag at all times...

    We progressed from that. We need strict rules, for autonomous cars now, but hopefully, we will look back at them as being draconian and/or silly as we are chauffeured around autonomously as routine, with huge benefits to society.

    1. JimC

      Re: Red flag laws ...

      By 1926 there were nearly 5,000 deaths a year on the roads. Abolishing that red flag law cost literally thousands of lives...

  7. Dave 126 Silver badge

    Another factor is urban planning:

    The road in question in this Uber case has a huge paved pedestrian walkway across its central reservation at that spot, with a tiny sign saying 'Do Not Cross'.

    Anyone who has seen a little book called The UK's 100 Most Stupid Cycle Paths will recognise this sloppy thinking.

    Whilst the jury is out on this case specially, and autonomous cars generally, I think there's scope for using their algorithms and data collection to identify and redesign dangerous road features by running simulations.

    1. handleoclast

      Re: Another factor is urban planning:

      The road in question in this Uber case has a huge paved pedestrian walkway across its central reservation at that spot, with a tiny sign saying 'Do Not Cross'.

      Stupid, but unsurprising given the stupidity of the average human.

      I think there's scope for using their algorithms and data collection to identify and redesign dangerous road features by running simulations.

      Actually, a better idea might be a shared (across all autonomous vehicle manufacturers) of accident blackspots. This is one case where autonomous vehicles could be far superior to humans. You might know of this problem location because you read about it, or observed it first-hand, or had a near-miss there, but your knowledge isn't transferred to many of your fellow meat-sacks. A lot of your fellow meat-sacks may drive through blissfully unaware of the problems that could easily arise. A shared db of problem areas, however, would give autonomous vehicles an edge over meat-sacks in those locations.

      1. John Brown (no body) Silver badge

        Re: Another factor is urban planning:

        "A shared db of problem areas, however, would give autonomous vehicles an edge over meat-sacks in those locations."

        Yes, this. I've said similar in the past. This NOT a pace for "walled gardens" such as Apple/Android/MS etc. Data relating to safety, road conditions, handling etc MUST be shared between AVs and all AV vehicles MUST be able to communicate with each other, no matter the manufacture or software supplier. And now, in these early days, is the time to make sure that they all work together.

        If it was left to the tech companies, it'd never happen. Too many egos. Traditional car makes may see the benefits. But there are so many players in the game now, we might just end up designing a camel.

  8. Bryan Hall

    FAIL

    Self-driving advocates say that because of all the sensors these vehicles have, that they are much safer than a human river. But in the released UBER video it's clear that the forward facing LIDAR and / or ultrasonic sensors would have easily "seen" this pedestrian before the headlights illuminated her - but the software completely failed to do anything with all that information. It had plenty of time to slow down and then drive to the left around her, but it did nothing. And given the "nothing to do" role for the human, he became completely inattentive and failed as a backup as he could have at least tried to swerve to the left once he could see her.

    PS - Either the camera video stinks or those headlights are useless at anything over about maybe 15 MPH. Better headlight technology is much easier (and cheaper) to implement.

    1. Andy france Silver badge

      Re: FAIL

      Having watched the video I would agree with you about the headlights being useless. Sadly given the very short time from the cyclist being visible to the accident I suspect that if I had been driving the outcome would have been no better.

    2. JimC

      Re: FAIL

      Have to agree. That should have been an easily avoidable collision. The pedestrian appears to have been crossing the road at a steady speed and direction. Its certainly possible for a pedestrian to do things stupid enough for it to be impossible for any reasonable automated car or human driver to avoid them, but this shouldn't have been one of them.

    3. Francis Boyle Silver badge

      Re: FAIL

      So Uber's system failed when it clearly shouldn't have. I think you're drawing the wrong conclusion here.

    4. Monty Cantsin

      Re: FAIL

      "Either the camera video stinks or those headlights are useless at anything over about maybe 15 MPH. Better headlight technology is much easier (and cheaper) to implement."

      I have a Volvo (an older model than the one in question,mind). The dipped beams are useless. The only disappointing part of the car.

  9. W Donelson

    Lawyers are delighted they have HUGE rich corporations to sue now, instead of poor you.

    Lawyers are delighted they have HUGE rich corporations to sue now, instead of poor you.

    1. LucreLout

      Re: Lawyers are delighted they have HUGE rich corporations to sue now, instead of poor you.

      Lawyers are delighted they have HUGE rich corporations to sue now, instead of poor you.

      They already do - for just 200 of my English pounds a year they get to sue some lowballing buffoon of an insurance company. I wouldn't insure me for 10x that and I know me!

  10. Oneman2Many

    Sending your phone location for P2V

    How many of you would be happy for your phone to broadcast its location to passing vehicles to improve your safety as a pedestrian ?

    Anonymously off course.....

  11. fishman

    Maybe it's just me

    But I always assume that cars will not stop when I cross the street, so I time it when there is a safe interval in the traffic.

    1. Anonymous Coward
      Anonymous Coward

      Re: Maybe it's just me

      That's a necessary assumption, and I do it too. But it only takes one time where you don't notice a car (if it is behind a bigger car so you don't see it and don't notice it is moving 2x as fast and will intercept your path) where even playing it safe like that will fail.

      A human can be "at fault" for running in front of a car, but a car and its driver (whether meatbag or bag o bolts) are still responsible for making every effort to stop. If I run out in front of you and leave you 1/20th of a second to react, you will not be judged at fault. If I run out in front of you and leave you 2 seconds to react, you will be judged at least partially at fault.

  12. ilmari

    From a northerly Scandinavian perspective, the bicyclist was out in the dark without wearing reflectors on her person, her bike was lacking basic side reflectors, and the legally mandated front light was not present or working at the required level. Crossing that road seems dubious at best, and I wonder how the bicyclist didn't notice the uber's headlights when, presumably, looking to both sides and listening for cars before crossing the road. The road could use fencing in the middle to prevent crossing by moose and pedestrians, except for designated moose and/or pedestrian crossings.

    The driver was distracted by presumably a phone, and speeding.

    Fine them both, improve the road, case closed?

    1. Richard 12 Silver badge

      The bicyclist is dead

      The vehicle doesn't seem to have even started braking by the end of the video.

      It should have done an emergency stop. A collision at 20mph is a lot more survivable than one at 48mph.

      It also clearly shows the basic fact that is bloody obvious to anyone with an even vague understanding of humans:

      The "safety driver" is useless, they don't pay much attention to the road because they aren't driving. You simply cannot concentrate on a task if you aren't actually doing it!

      1. John Brown (no body) Silver badge

        Re: The bicyclist is dead

        "The "safety driver" is useless, they don't pay much attention to the road because they aren't driving. You simply cannot concentrate on a task if you aren't actually doing it!"

        And that raises the question as to what exactly is in the laws allowing these experimental vehicles out on the road. In any normal car, the drive is "in charge" and therefore responsible. Do these new laws abrogate the "supervisor" in any way?

        1. Anonymous Coward
          Anonymous Coward

          Re: The bicyclist is dead

          Probably nothing much is in the laws, as some locales seem to be willing to risk road safety by letting companies do pre-alpha level testing with two ton projectiles. I guess they figure it will make their city seem "high tech", and somehow lead to high tech companies wanting to locate there?

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like