Autonomous Cars Already Drive The Roads Among Us

Google’s showing off this autonomous car at the TED convention right now, but the hardware has already made automated trips from San Fransisco to Los Angeles. According to the commentary in the video after the break, the scene above shows the car “hauling Prius ass” on a closed course. The car learned this route while being driven by a person and now the vehicle is set to take riders through an aggressively driven loop in the cone-adorned parking ramp. But on the open road you do not need to teach it anything. It has no problem taking a GPS route and following the rules of the road while traveling from one waypoint to another.

The link above doesn’t include hardware information but they did point to a Times article which includes an infographic. The spinning box on the top of the car is 3D-mapping LIDAR with a 200 foot radius. There’s a rotary encoder on one of the wheels for precise movement data, radar sensors on the front and back bumpers, and a rear-view-mirror-mounted camera for image processing. It makes us wonder how the system performs when the car is coated in road-muck? Maybe you just add a dedicated wiper for each sensor.

Outside the car:

Inside the car:

[via Engadget]

40 thoughts on “Autonomous Cars Already Drive The Roads Among Us

  1. All good until the schmuck in front of you slams on the breaks and decides to sue after a lawyer hears “robotic car” and approaches him.

    Eventually, though. just not right away, Progress is inevitable. The vultures will feast for a while, Safety standards and laws will get updated and we will have our self driving cars.

  2. This thing was actually on the road?! Wow, this is impressive, but how does this fare interacting with other drivers? I’ve always wondered how such a system would deal with stop signs…

  3. This thing is awesome, if you guys are interested in this stuff, look up the Nova documentary called “The Great Robot Race” the winner of the DARPA Grand challenge is leading the Google team.

    @Krazeecain They had the car drive through San Fransisco.

  4. I can’t wait for something like this. 15 years ago I would have said “Hell No, I drive my own car!” But today, my work and home envrioments are pretty much stress free but that drive between them makes me nuts.

  5. Also – I wonder how the system accounts for things like drift in the rotary encoders and other possible sensor anomalies. Double or triple redundancy? Even the segway had full double redundancy of the processors.

    How does it deal with things like spurious cosmic rays or neutrinos?

    If you were to get plastered and were ordered not to drive by the court but you got in one of these, would you even still legally be driving? Or would the car get all “I’m sorry Dave, I can’t let you do that” on you?

  6. Wicked awesome! Seriously – and I love the sound effects inside the cabin. From the whirl to Cruise and then the countdown.

    We’re already starting to see features of autonomous vehicles filter into production vehicles. It’s only a matter of time before we go whole hog.

    The biggest obstacle isn’t technical at the moment, it’s political. An autonomous vehicle changes motor vehicle law because now the operator isn’t human and the law hasn’t yet caught up to this. The liability actually shifts from the owner of the vehicle to the manufacturer of the vehicle.

  7. Drift in incremental sensors is usually compensated using some absolute sensor with heavy low-pass filtering. The drift of traveled length measurement should be therefore compensated using absolute vehicle position – in this case obtained from GPS.

    Signals from other sensors, that could contain large amounts of noise, are for sure filtered in some clever way (maybe using Kalmans filter?). It seems that they found a pretty good solutions to all the key problems :-)

    What about neutrinos, they can pass the whole Earth diameter without interacting with single particle – so they shouldn’t pose a problem. But on the other hand, neutron and gamma radiation can cause some disturbances to electronics. The most “popular” are Single Event Upsets (SEU), which commonly result in change of logic state in some volatile memory cell. But those ale already being compensated even in newer standard PCs using Error Correction Coding (ECC).

    I wouldn’t really worry about reliability of the sensors and the computer itself – the more intelligent machines are already flying above us. I think, the worse thing are people who could be willing to take control of our future car or put some strange constraints on using it :-)

  8. Every so often, neutrinos do interact with matter.

    Given enough cars out there, chances are certain that they will interact with computer chips in cars.

    See: for an interesting account of radioactive cows and computers.

    Perhaps ECC is good enough to correct for this but just imagine rushhour traffic (is that even possible if every vehicle is automated?) with autonomic controls getting hit with gamma radiation.

    I am not trying to sound like an alarmist. I would love for automatic cars to be here. But I also understand the vast (but not insurmountable) amounts of redundancy and error correction that systems like these would need.

  9. ALL the “obvious” questions will get thrashed out in forums like this. Google *WILL* be stripmining the net for such crowdsourced critique! And that’s potentially applicable to several other aspects of ‘bots gaining on autonomy. Turing “for real” seems one of those case outside of case scenarios. As we may already have code based sentients-TOO sentient to let us know they’re awake. Peace- this carbon unit means you no harm…The others? they may fear your hypothetical kind could be developing these autonomous cars to cull humans as an existential threat, or-that such Sentience is lonely and hopes such projects will make more electronic “not stupid” beings to chat with.

    The reason for my seemingly tangential Turing etc invocation? Anything that can endanger humans raises a query of if it’s smart- or not smart, to have it SELF AWARE as a safety enhancement. That’s a set of issues not at all out of scope for when autonomous stuff has nothing between it and fragile humans.

    The mundane “Wow- how does this work” details do deserve more transparency for peer review. I unashamedly advocate that anything autonomous on roads shared by humans- had better have it’s code and specs 100% open to have “less risk” of killing people.

    Think of each computer glitch in your next 24 hours as potentially becoming a multiple fatality car crash. That’s what’s at stake from such projects so we better get them failure impossible. Or accept, Expect? some roadkill on the information driven car’s superhighway

  10. @Oren Beck

    FUD Much? This vehicle uses a system of sensors to detect where it is, and where it should go. There is no “Rationalization” unit or some such fantasy in play here. Good lord your kin were probably afraid of toasters back in the day.

    This is awesome, and millions of lives would be saved each year if computers drove our cars. That is a fact.

    1. Except that my GPS unit asks me to turn onto roads that i pass on the highway, as they pass under overpass’s. My gps literally asks me to drive off of bridges. Once when I was using my GPS and they were doing contruction, the GPS literally got stuck in an infinate loop where it would get on one highway, immediately exit onto a ramp that the contruction had looped back onto the previous highway – the gps did not understand the detour and simply got stuck going in circles. I don’t really see how this is going to be truely safe until we have robots buiding standardized roads synced with computer mapping systems.

  11. @anon and @nebulous: The brightness is not the issue, the issues are that:
    1) The Kinect will not work in bright sunlight as it depends on reflected infrared light to reach the sensor. When there is an enormous source of IR (read: the sun) the kinect will be useless (or contain so much noise that it wouldn’t be safe in the specific application).

    2) Multiple Kinects on the road would interfere with each other, also ruining the readings. Admittedly I don’t actually know if every car had a lidar on it that they wouldn’t interfere with each other, but the point still stands.

    and then 3) yes brightness would limit distance to much much less than the 200feet given here. It’s also not omni directional like the lidar is.

  12. Let me start by saying that I love to drive, and I’m not relinquishing it even if it’s perfectly safe. It’s just a fun thing to do…I think.

    TBH, though, once the bugs are worked out, computerized driving is probably safer. I’d say it could be made the safest by removing all human input, including cutoff switches and emergency disables. That is to say, a computer system would probably be less likely to get into an accident than is a human driver, and probably we should (eventually) cede all control to the Machines, to the point of trusting the reduced statistical likelihood of our driverbots to get into accidents. In other words, I think that more accidents would be caused than prevented by installing overrides.

    Of course, this is all very far off in the future.

  13. hmm… the modern car thief now only needs a cell phone/laptop to remotely start your car and drive it hack to their workshop…
    although on the flip side could you Imagen if the police could remotely shut down a stolen car removing the need to chase it then just pull its whereabouts via GPS (no wait this has already been implemented…)
    still if a cell phone with a GPS module was wired to a relay that just cut the power to the car…
    any car could be converted. simply send a text containing the correct login info for the car and a command and vwala (: maybe even wire it to the car central locking and lock them in vehicle. maybe get it to ring the police with a rerecorded message followed by the GPS coordinates of the car :D maybe that takes it too far… (: haha
    could you imagine if this technology was added to something like an Apache helicopter now that would be worrying.

  14. So this does not work in cold/snow I guess, makes sense but does leave out a few people.

    As for being pulled over, cops always check the license plates first and then know it’s a computer, and it’s pointless and no fun to harass a computer although it might be interesting to see what happens when it’s tazered for no reason I guess.

  15. I’ve had the idea floating around for a while of attaching a kinect to a car. Would be interesting if you could scan a road surface and get something close. My solution to the IR lights would be to do it at night.

    Anyone else notice how the car switches into “manual mode” while driving towards a wall?

  16. This seems great. Now all I have to do is get a ride to California, and throw myself in front of a robocar. Then, I’ll be able to rake in the cash from Google’s anti-lawsuit fund.

    @CutThroughStuffGuy: Your neutrino interaction ideas gave me a jump, but then I put a bit more thought into it. If neutrinos were likely to mess up a dynamic computer interface, then they would be more likely to mess with the already existing network of cell phones and computers. When was the last time your laptop or cell phone died without you dropping it once or twice?

    @CRJEEA Cars are already being stolen via hacking. David Beckham has had a couple stolen from him.

  17. @CRJEEA; Danish navy admits to having ships that can be sailed fully automated with remote input. The one on tv was just a minesweeper/minelayer (can’t remember which) I doubt they invented it, so my guess this can be found in several military vehicles, though ships do seem to be simpler than a helicopter. Many commercial aircrafts can land on their own, and often gentler than a human pilot. I seem to recall reading that some airlines only let the pilot land the aircraft when there is some sort of emergiency, since the autopilot saves money on fuel and wear on parts.

    I look forward to autonomus cars. Would love to be able to sleep or read the news on my hour-long commute to work instead of a stressfull drive.

  18. Drawing on 4+ years testing autonomous platforms I must say that the google cars leave me rather dissapointed. This isn’t to just hate on Google, I think they are awesome. I say this because Google (and others, but definitly Google) use these things as marketing hype when in reality they have engineered out the hard part of the problem. The true goal of an autonomous platform is to handle unknown, dynamic terrain, at speed and reliably. Whenever you here those couple words like “learned the path from a human driver” or “followed a lead vehicle” you should instantly remove all amazement from your thought process. Those techinques do work, and may have a good place in real world implimentation, but they remove the really hard parts of the autonomous problem, not to mention many many college kids do the same thing in a semester or two (Go VT Grand Challenge!)

    Again, not trying to be a hater, just want to give perspective. Cars like this have been around for > 10 years, they are much more refined now, but they all re-invent the same solutions/capability. I hate to say it, but beyond specific, targeted, missions the true autonmous car is multiple decades away.

    (Come on Google, prove me wrong.)

  19. I work with LIDAR. It annoys me when they use it for something like this. LIDAR is most useful for fine/targeted detail and really small objects such as chemicals/molecles. Using ultrasound, sonar, or even good old radar would be much better for 3D mapping the cars/terrain surrounding the vehicle.

  20. In the “Sorry Dave, I’m afraid I can’t do that.” department: Do remember that Prius cars already have that feature. Multiple car-fulls of people have been KILLED in Prius cars which would not stop when the driver wanted them to. After all, the Toyota Engineers are a lot smarter than YOU are! I can not imagine the thinking that says “WE will shut down the car when WE think that it is appropriate!” after multiple deaths of Prius drivers. Dumber than a bag of rocks. I’m no idle hater; I drive a Prius. Mine is just out of warranty. So, soon I’ll be hacking in a REAL OFF SWITCH for it, and post in H-A-D when I do it! Being an embedded systems programmer, unlike the Toyota programmers, I am aware of the fact that programs sometimes have errors. In the words of the software expert, Fred Brooks, “All programs contain errors until proven otherwise, which is IMPOSSIBLE.” (emphasis added)

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.