Drones toss and catch inverted pendulum


Ever see flying robots doing stuff that you never suspected flying robots could do? I have.

First, a state estimator was used to accurately predict the pendulum’s motion while in flight. Unlike the ball used in the group’s earlier demonstration of quadrocopter juggling, the pendulum’s drag properties depend on its orientation. This means, among other things, that a pendulum in free fall will move sideways if oriented at an angle. Since experiments showed that this effect was quite large for the pendulum used, an estimator including a drag model of the pendulum was developed. This was important to accurately estimate the pendulum’s catching position.

Another task of the estimator was to determine when the pendulum was in free flight and when it was in contact with a quadrocopter. This was important to switch the quadrocopter’s behavior from hovering to balancing the pendulum.

Second, a fast trajectory generator was needed to quickly move the catching quadrocopter to the estimated catching position.

Third, a learning algorithm was implemented to correct for deviations from the theoretical models for two key events: A first correction term was learnt for the desired catching point of the pendulum. This allowed to capture systematic model errors of the throwing quadrocopter’s trajectory and the pendulum’s flight. A second correction term was learnt for the catching quadrocopter’s position. This allowed to capture systematic model errors of the catching quadrocopter’s rapid movement to the catching position.

Video: Throwing and catching an inverted pendulum – with quadrocopters | Robohub (Thanks, Kate!)



  1. Amazing! But it’s an inverted pendulum, not an inverted pyramid as the headline suggests. I’d like to see them do it with a pyramid, though.

  2. Neato.  And yet my terror grows.

    I know, I know.  The things aren’t out to get me.  But when the day dawns that they are, it turns out they’ll be able to outmaneuver and overwhelm me and my dwindling allies by a couple of orders of magnitude.

    On a totally unrelated note, are they EMP-resistant?  Flame-retardant?  Waterproof?

      1. Yeah, but all hilarity aside,  those ‘flame-retardant’ chemicals are not.   The whole thing was a scam perpetrated by the tobacco industries to pull attention away from people who died from smoking while falling asleep.  http://media.apps.chicagotribune.com/flames/index.html

    1. Yes, it does seem to. Dead reckoning would be stretching the capabilities of quadrocopters a bit.

      I’ll give them a year to work that out.

  3. I know it’s not the point… but these things have no actual knowledge of their absolute position. Their position is tracked with a mocap system. It doesn’t make it any less impressive… but it does make you realize just how crazy it would be get these things to be aware of their position without help.

        1. Seems if a Googlemobile can drive itself across a crowded continent without help, it won’t be too terribly long before a quadcopter drone will be delivering my pizza, photographing me in the bath, and chasing me down the stairwell and down the traffic-filled street.

          Even rocket science ain’t rocket science anymore.

          1. Google driverless cars don’t have to work with movements that total a centimeter in length.  They also don’t accelerate to 35 mph in a second and a half.  Additionally the cars can’t move in three dimensions, and they don’t invert their orientation through a 360 degree flip (in any normal circumstance).

            These little quad copters do all of that, and can play Peter Gunn the James Bond Theme to boot.

          2. Yeah, I get all that.  Is the point that these won’t be able to navigate an environment unassisted by external mocap equipment anytime soon?

            ‘Cause I don’t believe that.  When Mr Olsen said

            how crazy it would be get these things to be aware of their position without help.

            I assumed he meant difficult and prohibitively complicated.  Seems to me it’s just a question of miniaturizing and optimizing existing terrain-navigation algorithms and such.  No, I can’t do it.  But my ignorant ass predicts we’ll see completely self-contained dronebots that are fully aware of their position well before the next big election.

            Or maybe we won’t actually see them.

            (Never mind guys, I’m talking out that selfsame ignorant ass anyway.)

          3. @boingboing-096f32c997988c54d6d7c09ff0be4d32:disqus Oh, I’d say it’s inevitable that these little guys will eventually be able to do all these tricks autonomously.  But I don’t think it’s going to happen before the next election.  As far as I’ve seen, (and I could be wrong by now) these little helicopters aren’t doing any processing on their own.  They have a server rack doing all the motion capture, image and video analysis, and all the flight control and the learning algos to hone their accuracy.  And these copters are just the moving pieces they dedicate large amounts of hardware to control.

    1.  yeah – good point.  but maybe “in the field” you could designate three or four of your enumerable drones to serve as the external motion-capture references?

    1. When hasn’t calculus been cool?  All the cool kids these days are on their skateboards or throwing their footballs, or figuring out the lengths of shorelines (I had a shoreline-calculating-heavy textbook in high school calculus.)

  4. quadrocopter1: “Shall we decide the humans’ fate in a microsecond?”
    quadrocopter2: “Naw, let’s toss this thing around some more!”

    1. That’s how we’ll beat the machines, by distracting them with games of catch, just like they’ve distracted us with videos of cats.

Comments are closed.