Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×
AI Transportation United Kingdom Technology

Oxford Tests Self-Driving Cars 95

halls-of-valhalla writes "Using advances in 3D laser mapping technology, Oxford University has developed a car that is able to drive itself along familiar routes. This new self-driving automobile uses lasers and small cameras to memorize everyday trips such as the morning commute. This car is not dependant on GPS because this car is able to tell where it is by recognizing its surroundings. The intent is for this car to be capable of taking over the drive when on routes that it has traveled before. While being driven, the car is capable of developing a 3D model of its environment and learning routes. When driving a particular journey a second time, an iPad on the dashboard informs the driver that it is capable of taking over and finishing the drive. The driver can then touch the screen and the car shifts to 'auto drive' mode. The driver can reclaim control of the car at any time by simply tapping the brakes."
This discussion has been archived. No new comments can be posted.

Oxford Tests Self-Driving Cars

Comments Filter:
  • by lophophore ( 4087 ) on Monday February 18, 2013 @08:54AM (#42934675) Homepage

    on your car, you have "brakes". if the brakes break, then you have big problems.

    Kindly consult the Oxford English Dictionary.

  • by Anonymous Coward
    For a prototype it's OK but I wouldn't bet my life on anything running on consumer-grade devices.
    • For a prototype it's OK

      Actually it's a great idea. If the car crashes, you can just blame the driver for holding the ipad wrong.

      Actually, the reason they have an ipad in the dashboard is almost certainly because a student thought it would be way cooler to have an ipad with an app relaying data over some connection than plugging a small screen into an embedded PC.

    • I would expect that even in the prototype, the iPad is just being used as an interface to the control system making the actual decisions. iPad crash = no problem: you can still get out of auto drive by applying the brakes. And for a prototype, a tablet makes a cost effective and feature rich user interface (in a production system you'd probably use a custom device integrated into the dashboard).
  • ... a shadowy flight into the dangerous world of a man who does not exist...
  • "breaks" = "brakes" (Score:3, Informative)

    by nukenerd ( 172703 ) on Monday February 18, 2013 @08:58AM (#42934705)
    "breaks" for vehicle brakes keeps coming up here. Is this an Americanism?

    But this is a story about Oxford FFS, the cultural heart of the English language, UK version.

    http://en.wikipedia.org/wiki/Oxford_English_Dictionary [wikipedia.org]
  • Don't have one to check.

    What if I need to accelerate to avoid a hazard? How does "braking", which I assume the poster meant but does not understand, help me regain control?

    I've heard that "cruise control" systems allow temporary acceleration, then fall back to the desired speed, while braking puts them into a sort of standby mode which can be resumed. Maybe they should consider something similar.

    • You're correct on the cruise control. Here's how it works for every car I've driven that has it:
      - Accelerating by pushing the pedal allows you to speed up. Releasing the accelerator allows the car to slow back to it's previously set speed
      - Braking will automatically unset cruise control but CC will remain on. You can then hit Resume to have it go back to it's previously set speed

      It's a pretty good model that's simple and natural enough for people to learn it in about 2 sentences. If you have to accelerate,

      • by AVee ( 557523 )
        It's mainly a good system because the car responds to user input exactly the same regardless of the cruise control. This makes it easy to explain, but more importantly it means all those reflexes you've developed will still be useful in an emergency situation. When shit happens on the road you don't want your response to be different depending on the state of the cruise control.

        As a side note, I've found my car also unsets the cruise control when the ESP gets activated (which makes sense) and when acceler
      • I'm thinking of this as Cruise Control 2.0

        It's a nice transition to autonomous cars. The built in redundancy is a human behind the wheel. The advantage is that the human won't have to assume control every time the car comes up on a slower moving vehicle and it could work on surface streets or in commuter traffic.

        The learnings from this will only improve the guidance systems for autonomous driving.

  • by Chonnawonga ( 1025364 ) on Monday February 18, 2013 @09:10AM (#42934785)

    You'd think it would be obvious to the folks at Oxford: if you're building 3D maps, and storage is getting relatively cheap, why not just build 3D maps of whole regions so the car knows its way around? Then the human can pick any route, rather than having to teach the computer.

    • by serviscope_minor ( 664417 ) on Monday February 18, 2013 @09:25AM (#42934923) Journal

      You'd think it would be obvious to some of the folks at slashdot that pontificating about a grand idea is much, much easier than making a simpler idea actually work right.

      • We're talking about pre-loading a cache, or pre-computing. This is NOT a grand idea. It's a pretty common thing.

    • by Xugumad ( 39311 )

      Time.

      Despite what is frequently suggested, there's a much greater scarcity in skilled developers/researchers/whatever than of ideas for them to spend time doing.

      • If it's about time then it's also about money and security. If there were money in fun University research projects like this, then I wouldn't be working for an engineering firm. Students don't really make money..

    • by AlecC ( 512609 )

      Because the pictures produced by 3-D maps look much less like the real world than the real world. While the 3D map will probably tell you the height and approximate colour of the building, the real world contains ad-hoc signage, discolouration, texture, minor details such as lamps, signs and street furnityre which will be of great assistance in fine location. I would rather a car that was able to drive down the real road that it had driven down before than one which could drive down a computer model of a ro

  • So an iPad is telling me it knows the way and is able to guide me through Oxford traffic safeley? Erm, not thank you.
  • by Anonymous Coward

    I like how tapping the brakes gives control back, it's like complete cruise control. Wonder what happens when it snows 3 ft (guess what part of the country i live in) or a different situation where the landscape changes. Even after the roads are clear there are still walls of snow that would block any sensors abilities. Maybe we need something up on the phone poles or in the center lane embedded that tells cars where they should go. Seems more reliable that way, but I guess that would be more public in natu

    • If we're 3 ft deep in snow then I think I'd prefer to take control myself, thank you. Indeed it applies to any situation where the environment is not what the computer might be expecting.

      I like the approach here though, the human has to teach the computer once but then the computer can contribute as & when needed. I can take control quickly if I need to.
  • So, what happens if, say, a building gets demolished, or a set of trees are cut, or it snows? If the software is looking for specific topography, which it learned from previous trips, isn't it as likely as humans to get lost when things change?
    • by AlecC ( 512609 )

      One would sincerely hope that it is depending on more than a single point, or single building, fix. It should have at least two or three major reference points in use at any moment, plus more ahead that it is acquiring for future use and more behind that it has passed but have not gone out of sight. I would hope that at any instant it has at least twice as many reference points as it needs.

      And if all else fails, just like a human, it should have enough absolute road sense to come smoothly to a halt however

      • I agree, but a major construction project could quickly exceed the count. What my (warped) imagination sees is, pull out the drive on the way to work, tell the car to go to work. Now that you're not driving, the day's paper comes out, but half way into the sports section, the car announces it's confused and you should take over, whilst zipping down the road. Or any number of other similar situations. The point is, if there's *any* possibility of the program suddenly demanding a driver, it largely negate
        • by ShanghaiBill ( 739463 ) * on Monday February 18, 2013 @10:53AM (#42935807)

          the car announces it's confused and you should take over, whilst zipping down the road.

          Unless the people developing this are complete morons, there is no way this could happen. The car knows its safe braking distance, and if it cannot map out a route beyond that distance with an acceptable degree of confidence, it would pull over to the side of the road, come to a stop, and then alert the driver.

          • Or, keep driving until you can take over, but this doesn't change my point. There's also the problem of the AI has only learned what you've taught it - missing a turn (for whatever reason) could leave it on roadways it's never seen. I'm fairly certain the coders aren't morons, but don't believe a 'learning' program would benefit most drivers. It would demand much more oversight than most users are willing to give.
            • Or, keep driving until you can take over

              This would require not just an extremely stupid and incompetent programmer, but also a complete lack of any code review, and no system testing whatsoever. It is conceivable that some stupid people are accepted by Oxford, but exceedingly unlikely that they could comprise an entire team of developers. Do you also worry that buildings might collapse because the architect forgot to specify mortar between the bricks, and nobody noticed?

  • by Joe_Dragon ( 2206452 ) on Monday February 18, 2013 @09:25AM (#42934927)

    can it change lanes? can it route around road blocks? can it stop for red lights with out getting messed up?

    • Re: (Score:3, Insightful)

      by AlecC ( 512609 )

      To 1 and 3: almost certainly. To 2: probably not unless it has already "learned" the alternative route.On the other hand, one difference between computers and humans is that you can copy the "learning" from one computer in a way you cannot copy from one brain to another. So it would not strike me as unreasonable for a net-connected car to download the images of a detour route within a few seconds of recognising a roadblock.

      • by tftp ( 111690 )

        So it would not strike me as unreasonable for a net-connected car to download the images of a detour route within a few seconds of recognising a roadblock.

        Especially if the workers have a small beacon that broadcasts at, say, 5 GHz the map of the area that shows how to drive around the work area. All cars in vicinity receive that and can act upon these instructions if they are signed and the chain of trust is good enough.

        That would be better than what we have now - a mass of cars trying to get by the w

  • Theoretically this mixed with a service that has surveyed the entire road network could be amazing. It would remove the need to have travelled the route previously.

    Although I do wonder how it copes with changes in road layout/diversions etc.

    • If we could get pass being uncomfortable with letting go of some privacy, this could be amazing as it could supplement street view and update it in "real time". Imagine a road becoming under construction. It is would update in Google Maps and thus reroute other drivers. We could also see this construction on Street View as well. And if we export these cars to N. Korea...

    • Always wondered what Google's motives were for their huge investment in Street View...

      Note that much of the Street View data already includes a 3D aspect and they have a very basic idea of the location of building surfaces.
  • TMI Siri (Score:5, Funny)

    by wynterwynd ( 265580 ) on Monday February 18, 2013 @09:40AM (#42935055)

    "I see you are driving down Laurel Ave and I can drive you to your destination. Are you heading to:

    - Bosco's Liquor Store (1.73 mi)?
    - The Bouncing Pasty Gentleman's Parlor (2.64 mi)?
    - The Purple Nurple Tobacco Accessory Shop (1.25 mi)?"

    ".... Siri, change profile to 'Mom'."

    "Okay. Changing user profile settings to 'Mom', please wait"

  • and yet.... (Score:5, Funny)

    by duckintheface ( 710137 ) on Monday February 18, 2013 @09:40AM (#42935059)
    All that time and engineering effort. All that programming expertise. And yet.... it still drives on the wrong side of the road.
    • Re: (Score:3, Informative)

      by xorsyst ( 1279232 )

      No, it drives on the right side of the road. Which is to say the left side. Not the right side, as that's the wrong side. Clear?

  • by Anonymous Coward

    Who writes these summaries?

  • Hey, I've been down this road 200 times, and you've went three different ways! Want me to drive the rest of the way? All I did was RTFS, but that's a problem with bad summaries...unless that's what the article implies. How does it know which of the three ways you want to go? That sort of thing could kill this innovative idea if it's indeed the case.
  • ...till the first foggy day.

  • don't drive past The Mirror Store.

"Being against torture ought to be sort of a multipartisan thing." -- Karl Lehenbauer, as amended by Jeff Daiell, a Libertarian

Working...