What Google's self-driving car sees

Charlie Warzel: "THIS is what google's self driving car can see. So basically this thing is going to destroy us all." [via Matt Buchanan]


  1. I’m not sure what I’m supposed to be noticing. It looks like the car can see a lot. Is it the blind spots right around the car?

  2. Now compare that with the limited visibility of a human driver.  The human would have about a 2-degree cone of ultra-sharp foveal vision, surrounded by another 8-degree cone of semi-blurry para-foveal vision, surrounded by perhaps another 30-degree blur of peripheral vision.  I suspect given a choice, we’d prefer the 360-degree perception of the self-driving car.

  3. There’s a video of a lecture by one of the Google car developers (in more  of a classroom setting) where it shows video of the car driving along with the corresponding animated 3d representation of what the car thinks is out there (an animated version of the image above).

    They also explain how the car logs the position of objects every time it goes down the same route and can learn that things that are there every time are things like mailboxes or lampposts while other things that aren’t moving but are not always there or in the same sport might be parked cars – and it treats the knowledge accordingly.

    They also talked about how they programmed it to obey driving laws such as 4-way stop etiquette (including waiting what, 3 seconds after stopping?) and how other real drivers didn’t follow the law and they had to adjust the programming accordingly.

    They also show a video of the car turning left at a 4-way stop and reacting to a pedestrian who appears in the crosswalk without warning and how it reacts perfectly, yielding.

    1.  “Pedestrian who appears without warning” is funny.  I know when I’m crossing the street, I always loudly warn everyone in the vicinity that I am indeed stepping into the crosswalk.

  4. The dark blue information layer appears to be radar; you can clearly see interior walls on the left.

    If this assumption is correct, just imagine being downtown, and 50 nearby active cars are transmitting radar. Imagine the howl from those who oppose the radio signals from cell phones and Wi-Fi. In this case they might have a point:  That radio energy is going to add up quickly.

    Likewise, the very brief (and invisible) several-times-a-second laser flash from the car’s navigation system hitting your eyes likely isn’t a problem.  But once you have scores of nearby cars all doing the same?

    I’d also be interested to see whether the radar and laser signals from dozens of cars operating in close proximity will interfere with each other.

    1. Actually, if they design this system really well, cars might actually be able to share information and reduce the quantity of probes they have to send out. Could also have traffic signals tie into the sharing network, and maybe even smartphones so that pedestrians could register their location. Even better, if everyone is using GPS navigation they could share the route they plan to take, and cars could automatically make room for each other.

      In a perfect world anyways. In the real world someone would hack the system in about 30 seconds to publish false info and cause a bunch of problems.  And law enforcement would mandate access to disable vehicles.

      But I can dream, right?

    2. It’s not radar. It’s probably LIDAR. Radar doesn’t provide anywhere near the resolution of detail that this car needs. The blue/white stuff is probably dead space behind refracting windows.

      1. The yellow data layer appears to be LIDAR, complete with the occasional interior wall seen through a window.  They match up nicely with walls in the blue layer.

        The blue layer however also shows walls BEHIND walls.  LIDAR won’t do that.  RADAR will.

  5. by “…this thing is going to destroy us all” does he somehow mean dramatically reduce the 1,240,000 traffic deaths worldwide? 

    1. Sure it starts out that way, but then the cars become sentient and homicidal, their reign of terror only brought to an end because they can’t refuel themselves and eventually run out of gas.

  6. The system seems great — now look what happens when you put it into manual (i.e. human) mode:

  7. Bring it on.  While taxis are an obvious first choice (with prepay an obvious necessity), I think transport trucking will be a very close second.   In many things I don’t trust computers, but in driving I really don’t trust humans – we have too much of a tendency to ignore dangers and assume the best. 

    I have been driving for 25 years without an accident, but at least a part of that is sheer luck – there have been moments where I wasn’t paying attention and it was only chance that nobody was hurt.  A computer never stops paying attention, and our traffic fatalities would drop enormously.

    Of course, media cycles being what they are, the first accident with a self-driving car will be played up as a huge menace.

Comments are closed.