ETH's sFly quadcopters navigate with stereo cameras, not GPS

From IEEE Spectrum:

Quadrotors are famous for being able to pull all sorts of crazy stunts, but inevitably, somewhere in the background of the amazing video footage of said crazy stunts you'll notice the baleful red glow of a Vicon motion tracking system. Now, we don't want to call this cheating or anything, but we're certainly looking forward to the day when quadrotors can do this outside of a lab, and the sFly project is helping to make this happen.

What makes the sFly project, led by ETH Zurich's Autonomous Systems Lab, different is that the sFly quadrotors don't rely on motion capture systems. They also don't rely on GPS, remote control, radio beacons, laser rangefinders, frantically waving undergrads, or anything else. The only thing that sFly has to go on is an IMU and an onboard camera (and an integrated computer), but using just those systems (and a "very efficient onboard inertial-aided visual simultaneous localization and mapping algorithm"), sFly is capable of navigating all by itself. And if you have a fleet of sFly quadrotors, you can use them to make cooperative 3D maps of the environment:

Each quadrotor is completely autonomous, but they're also equipped with two extra cameras that stream stereo imagery back to a central computer over GSM or Wi-Fi that takes the data from several quadrotors and combines it into an overall 3D model of the environment as a whole. Then, the computer can guide each robot to an optimal surveillance site. The idea here is that you'd be able to rapidly deploy an sFly system with a swarm autonomous quadrotors in a disaster area or somewhere else without any infrastructure (or even a GPS signal) and still be able to take advantage of some clever autonomous aerial mapping.

Views: 1680

Comment by Veikko Vierola on May 3, 2012 at 11:19am

I wonder if it could be possible to put ortophoto (from Google Earth for example) in to autopilot memory and then let autopilot navigate by comparing the original photo to the realtime image below??

Comment by Kevin Bouchard on May 3, 2012 at 11:30am

Title should read "stereo cameras", no?

Comment by Veikko Vierola on May 3, 2012 at 11:38am

stereo i presume

Comment by Jack Crossfire on May 3, 2012 at 12:49pm

In last week's video, http://www.youtube.com/watch?v=AdmTGAo1Jvg

they showed a 1st step that involved manually flying over the area to build a 3D map, then switching on the vision to automatically fly over the same area.  It wasn't obvious how the autopilot mattered if it had to be flown manually to build the map.

Comment by Jonathan Lussier on May 3, 2012 at 1:34pm

How long until the quadcopters are stable and powerful enough to lift a Kinect + Atom board?

Comment by John Leichty on May 3, 2012 at 6:33pm

Jonathan, about -1.5 years. :-)
http://www.youtube.com/watch?v=eWmVrfjDCyw 

Comment by Ellison Chan on May 3, 2012 at 6:50pm

Doubtful a Kinect will work outdoors.  It projects an infrared grid, and only has a certain range.

Comment by Cliff-E on May 3, 2012 at 7:50pm

Kinect outdoors is a no go. Tried one for a UGV--no good.

AR/scene reconstruction is a very promising tech for nav, but I wonder about the proessing power and latency (if a GCS does the processing) challenges when navigating urban-jungles and having multiple aircraft to control--which is what their use case appears to be.

Comment by Michael Zaffuto on May 4, 2012 at 2:41am

Flash LADAR is a big boys outdoor capable "kinect" technology for 3 dimensional imaging,

 

http://advancedscientificconcepts.com/products/portable.html

Small and lightweight enough for an UGV or reasonably sized RPV,

not small or cheap enough for hobby quads...yet!

 

Comment

You need to be a member of DIY Drones to add comments!

Join DIY Drones

Social Networking

Contests

Season Two of the Trust Time Trial (T3) Contest has now begun. The fourth round is an accuracy round for multicopters, which requires contestants to fly a cube. The deadline is April 14th.

A list of all T3 contests is here

Advertisement

© 2013   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service