DLR Crash Report: Towards a Standard Crash-Testing Protocol for Robot Safety

Tags: , ,

15 Responses:

  1. humble_fool says:

    I just realized that slow-motion lab footage looks like surrealist chiaroscuro paintings.

  2. duskwuff says:

    "... I shove around the blind people."

  3. seminiferous says:

    Wow, even the crash test dummies have to wear closed toe shoes. And perfectly shined Oxfords no less.

  4. gryazi says:

    1. Ow.

    2. While it's cool, isn't the feedback response in the controlled case kind of overengineered for industry? Just sheathe the whole thing in your favorite choice of touchscreen material and cut the power when it bumps something. For parts with serious inertia you might want more, but on an assembly line you wouldn't want the thing recoiling only to smack into whoever's running over to help. (Although, after implementing that, I guess the more complicated feedback is still handy for exposed tool surfaces -- if not on work surface && spinning drill bumps what might be flesh, stop before you make it worse.)

    3. Sometimes I wonder what the world would look like if the concern for safety took precedence over getting-things-done-for-whatever-reason. (You don't have to imagine a totally paralyzed Puppeteer society for this, just the engineering equivalent of the Hippocratic oath.) Roads, in particular - if thwacking into another person/deer/family_pet were just never acceptable from the outset, what would we have come up with by now? A bunch of really low zip-lines?

    • strspn says:

      I wonder what the warning stickers on predator drones say.

      In other news, the pentagon turns to brain implants.

    • skreidle says:

      Just sheathe the whole thing in your favorite choice of touchscreen material and cut the power when it bumps something.

      a) Unbelievably expensive.
      b) Too sensitive to false positives.
      c) Not nearly quick enough to high-inertia devices (as most industrial equipment is.) To analogize, would you rather get hit with a baseball bat wrapped in touchscreen, or one wrapped in foam padding with sensors in it?

      • andr00 says:

        d) oops, it just squashed you with the (sensorless) car door frame it was holding.

        • skreidle says:

          Ding ding ding! That too.

          (I work with some industrial robotics--specifically, Bosch manufacturing robots equipped to handle large tape cartridges--and the gripper head (which is a fine combo of tough and delicate) has springs and sensors behind it, in case it crashes into something rather than seating the tape in a drive or storage slot properly. Now, this particular setup is not designed to protect against human interaction; there are e-stops built into the doors so they can't be opened without powering down the robot, but those are often cheated open in practice. However, same concept holds--touch sensors wouldn't work in this scenario, either, and it's a pretty simple one.)

          • gryazi says:

            Point taken; I was thinking along the lines of a sawstop approach versus the BoingBoing cuteness of a saw that skitters across the room and hides when it nicks you, however it gets implemented (and having a really minimal full-body/full-massive-moving-armature contact sensor, of a sort not likely to be hair-triggered - membrane, then? - would seem like a means of accomplishing that.

            Sawstop is a rather bad example because it's self-destructive - in an industrial setting it'd be worth investing in rapid restart ability - but it's another example of that 'what is worth more, getting work done or not losing a finger?'. And I would wish that grippers for Heavy Dangerous Objects were designed such that Bob tripping over the power cable doesn't cause the bot to drop the 55 gallon drum of nitroglycerin on the floor.

            [The Predators are Mostly Harmless, it's their payload that needs the disclaimer...]

          • lionsphil says:

            Well, putting your arm in the way of the thing while it's stabbing a knife at you is certainly a vote of confidence, even if he's able to just move it downwards to match.

            I'm actually surprised that without the arm being a firm obstacle it still stops.

            • strspn says:

              Multiple sensors (and multiple sensor modes, i.e., pressure and optical both) can do far better on both false positives and false negatives. Until Joe the machine learning enthusiast in accounting loads a cost-benefit analysis program into the system, and the robots decide they need to calibrate with more data on how much it would cost if they chopped off your arm.