Follow Slashdot blog updates by subscribing to our blog RSS feed


Forgot your password?
AI Science Technology

Flying Robots Flip, Swarm and Move In Formation At UPenn 122

techgeek0279 writes "The University of Pennsylvania's General Robotics, Automation, Sensing and Perception (GRASP) Laboratory has released a video of flying nano quadrotor robots. Inspired by swarming habits in nature, these agile robots avoid obstructions and perform complex maneuvers as a group."

This discussion has been archived. No new comments can be posted.

Flying Robots Flip, Swarm and Move In Formation At UPenn

Comments Filter:
  • by Gabrill ( 556503 ) on Thursday February 02, 2012 @09:24PM (#38910771)

    Formation does not equal swarm. A swarm of insects doesn't have a known predetermined formation, nor does a flock of birds (not talking about duck v's). Impressive flight characteristics and preprogrammed flight formations, but I don't see anything that suggests you can tell it a destination in the wild and the group will be able to navigate there around random trees, buildings and other obstacles. For example the brick wall pass did not need the whole swarm to pass through the one window. A natural swarm would have flowed around as well as through, because each member would make an effectively random choice about which path to take.

  • by John Hasler ( 414242 ) on Thursday February 02, 2012 @09:58PM (#38910965) Homepage

    This is not a swarm of robots cooperating. It's a single computer remotely operating a bunch of quadrotors. Impressive, but not what you imply that it is.

  • Hey all - These guys work down the hall from me. I don't work with them, but I've seen the lab.

    Basically, it seems like it's a motion capture setup with IR cameras and some mostly off-the-shelf software to track 3D position (standard mocap stuff, which I have worked with). I think each drone has an IR emitter on it (you can see it in some shots since the camera has no IR filter). The novel thing here is the algorithmic work required in keeping track of each drone and planning out all the trajectories relative to the other bots (see the figure 8 demo at the end).

    It's not going to fly through your window any time soon, unless you can fit a Kinect and some serious horsepower on there without going over the weight budget. But there's no reason to think that the algorithms wouldn't work to control the local bot, with some sort of ad-hoc mesh network for the synchronization.

  • by Kozz ( 7764 ) on Thursday February 02, 2012 @10:29PM (#38911129)

    I think what the GP was trying to say was that it wasn't displaying what is sometimes called "emergent behavior". In this kind of tech, when we discuss "swarm" behavior, we're usually talking about individual entities that don't have very many rulesets except for things like "don't hit your neighbor", "don't hit obstacles", and "match your neighbor's approximate direction and velocity". You can see this in swarms of insects or birds (for example), and of course they're not communicating with each other on their planned trajectories, but the emergent behavior is fascinating.

    (disclaimer: I'm no expert in this field, I just read lots of slashdot and others. someone will pipe up and correct my mistakes, which I welcome)

  • by recharged95 ( 782975 ) on Friday February 03, 2012 @01:41AM (#38912109) Journal

    Bingo, they are presenting coordinated motion instead. They are close to a swarm, they are independent, but not sure if they are still commanded by a central computer (off-board), which means it's not a swarm by a mile. In hindsight, if they are playing back a script on-board each copter, it would be considered modeling swarm formation, but nothing close to flocking (there needs to be a leader quad).

  • by TheLink ( 130905 ) on Friday February 03, 2012 @09:25AM (#38913853) Journal

    These flying bots remind me of you average Alaskan mosquito.

    Those bots are not even half as advanced as a mosquito (and far from houseflies) though. Mosquitoes can fly for one to four HOURS: []

    Mosquitoes can navigate and orient in dynamic environments without requiring external cameras and computers ( [] ). They can find their own sources of fuel, and avoid active and passive threats. They can even produce new mosquitoes in a few days/weeks without a factory.

    They can get confused by bright/UV lights, but it's still quite impressive considering their brains are so tiny.

    So these bots are interesting, but there's plenty of room for improvements :). We're still not in danger of Skynet bots yet...

panic: kernel trap (ignored)