Hilarious DARPA Robots Falling Video

If you’re worried about Skynet, take a quick gander at the outtakes reel (YouTube, inlined below the break) from the DARPA Robotics Challenge (DRC) and you’ll feel a lot better. The IEEE Spectrum reporters assigned to the DRC took a break from their otherwise serious coverage and made this funny compilation of all the multi-million dollar robotic fails to go along with this article.

robot_fallingIt’s hard not to empathize with the robots, and we’ll admit that we winced a little bit with the first couple falls. (But after three or four, started breaking out in maniacal laughter.)

So you’re probably safe from the Terminator for at least a couple of years. But before you start getting too smug, human, remember what happened with the DARPA Grand Challenge that focused on autonomous cars; no teams finished in 2004, but five teams finished just a year later in a reasonably tight race. The DARPA Challenges are as much about learning as anything else.

And bear in mind, as you watch these robots falling over while trying to turn a doorknob, that they’ve sat down in and driven a car in a previous phase of the trial. That’s pretty darn cool.

46 thoughts on “Hilarious DARPA Robots Falling Video

  1. It’s weird how many of them failed so dismally when you think this is a challenge where the makers prepped it to their max ability. And frankly I expect more at this moment in time. I think this should not be happening to more than one unit, or maybe two tops on a bad day.

    1. Watch out for the badass! :O
      The weird thing I see here is the crazy amount of mr. knows-it-all in the comments section! Even in an amusing video, people come to brag about how technology is not corresponding to their super engineering standards.

      Its a challenge, probably tested in different environments than lab, doing complex (for robots) actions. Its normal that they fail, the purpose is just to have a laugh at it. It happens all the time in regular robot competitions let alone in a DARPA challenge, teams sure have more money constraints than DARPA itself.

      “I think this should not be happening to more than one unit, or maybe two tops on a bad day”
      You think? On what grounds? Did you come from the future with ultra knowledge on the matter, Mr. Terminator?

      1. You think it’s expected to see a state-of-the-art robot fall over while it’s standing still on a flat surface?
        You think expecting it not to is the equivalent to expecting to see terminator and such movie characters?

        I think that if engineering and technology is really in the state you suggest we should all avoid escalators and elevators (AKA lifts). And don’t even think about getting into an airplane or car.

        1. Not quite.
          Do you have any idea how many years and man-hours were pumped into avionics and automotive systems to make them as reliable as they currently are?
          The kind of robotics on display at the DARPA challenge is relatively new and even builds off much off the tech present in the aforementioned industries. It cannot be expected, at this stage, to “just work”.
          Your concerns would be legitimate in 5 or so years, but certainly not now.

          1. So you think it’ll take another 20 years for a robot on a flat surface to not fall over? Well I have news for you, we already have not-toppling technology for a while now.
            And it’s not like all those robots have dainty little human-like feet, we are talking about robots with large flat surfaces. And I can certainly see them fall once in a while when they do one of those rocks track walks, that is to be expected, but for them to flip over or fold over just standing there means you have severe software and possibly sensor bugs that have little excuse to exist past the alpha stage.

            And I think that if someone released a software package with such bugs you and everybody else would complain and a class action would follow to get your money back. As seen with recent cases where they released extremely buggy very early beta games as a marketable product.

          2. I’m sure you can do better ;)

            Now you’re just arguing for the sake of it. You clearly have no significant experience with this field of robotics, or robotics in general, so why don’t you just listen to people that do mate.

            FYI… you are looking at “products” in their alpha stage. These are experiments. Initial forays into a promising field of robotics.

          3. DItto – I’m thinking you’ve never actually gone through a real engineering project yourself before. Wouldn’t complain so much about them failing. No one said these are final stage commercial products.

            Anywho, either way, these are /alpha/ build robots. Also, most of these teams are at universities, composed of graduate (maybe even undergraduate) students who are indeed doing a lot of this work last minute (as there is a fixed, unrealistic yearly deadline for this challenge) , and took on the project as an aside from their primary academic tracks. Your current state of the art for advanced feedback systems with military-grade reliable mechanics is probably going to be the stuff from Boston Dynamics, and they have their share of outtakes as well. It turns out it’s pretty difficult to design the feedback loops to mimic human motions with current macroscopic weight/size mechanical components, and it’s a very large, difficult, multidisciplinary, coordinated engineering project.

        2. Realistic bipedal movement in all environments with out a tether and no himan operator is.. To say the least, on the edge of impossible by current standards.
          The fact that these things can MOSTLY manage to do it is frankly amazing.

    2. These are cherry-picked cut scenes from probably many of hours of video that was shot at the challenge. How many of them really did fail this dismally and this often? I’m not sure how many robots were at the challenge.

      1. no this was the normal, all but one team fell over (most multiple times) and only one team that fell even tried to have their bot get up by itself while most others rest the robot manually.

    3. I believe the big problem is the makers didn’t compensate for the weights they forced all their robots to wear like a huge backpack. First, our brains are our controllers and the don’t make us top heavy. Second when we do wear weights on our backs, our core body muscles tighten to hold us upright. This a funny vid!

    4. I mean you just copy humans and build “tendons and stuff” out of metal, stupid scientists and their trying to do something i can do in my head easily.. Please see Whatnot’s new balancing robot that never falls down in the following link:

    5. None of them were ready for prime time. They weren’t even ready to leave the lab let alone compete in a contest. So it turned into a joke fest.

      Bad thinking and non-existent proof of concept testing all around. I wouldn’t let those crews touch anything life critical, that’s for sure.

  2. Looks like falling over is fairly common among robots and it’s great that they can sit to drive cos it means they can control a fall enough to not look stupid every time. I wonder how they cope with getting up again after they’ve fallen, that would be a much more interesting video, although I’m guessing there’s lots of flailing and humans involved.

  3. Of course the video is supposed to be funny (and it is)!

    But every one of those mishaps are due to a miscalculation of center of gravity. I wonder, if there’s a way to dynamically, and quickly calculate it, along with multiple accelerometer chips?

    I’ve played around some with ROS, and haven’t noticed much discussions about dynamic CoG calculations. Is it happening? I’m sure it is.

    For experienced robotics physicists, does this sound like a good plan? If not, how do you solve this problem now?

Leave a Reply to WhatnotCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

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