Def cool!
It would be awesome to fully automate it, eg using "enough" vertical acceleration (and assuming one was flying smoothly/hovering before) to trigger the right behaviour. How accurate is the attitude estimate during the flip?
Ralph wrote:For anyone who also finds this cool, here is a short video showing some flips... made on a typical Friday afternoon at the office
Very easy to implement and quite fun and reliable! All flips in the clip are performed after each other; I didn't have to cut out any fails. I even got a double flip working a few times, though it doesn't quite flip fast enough to get this stable.
I hope to work on coding some flipping intelligence on the crazyflie itself in the next few weeks. Will post the results if anyone is interested in it.
omwdunkley wrote:Def cool!
It would be awesome to fully automate it, eg using "enough" vertical acceleration (and assuming one was flying smoothly/hovering before) to trigger the right behaviour. How accurate is the attitude estimate during the flip?
I'm not sure yet. I'm now finishing up a basic flip purely using timers. If I've got that smoothed out a bit, I'll plot some graphs with the gyro/accel/baro data during the flip. Then I can see what data I can use to perfect it further.
Thats pretty much how I did the crazyflie free fall competition too..drop it lots, export data to matlab, and test different classifying functions that would trigger at (and only at) the right time fast enough ) That combined with a mini state machine (detecing successive "features" in the right order) worked really well. Good luck! Looking forward to seeing what you can come up with!
Im pretty sure the baro will be useless. Im going to experiment using a second crazyflie as a barometric reference. Hopefully that might improve the barometer reliability.
Keep us posted!
Ralph wrote:
omwdunkley wrote:Def cool!
It would be awesome to fully automate it, eg using "enough" vertical acceleration (and assuming one was flying smoothly/hovering before) to trigger the right behavior. How accurate is the attitude estimate during the flip?
I'm not sure yet. I'm now finishing up a basic flip purely using timers. If I've got that smoothed out a bit, I'll plot some graphs with the gyro/accel/baro data during the flip. Then I can see what data I can use to perfect it further.
tobias wrote:The flipping is just beautiful! So the path is fully controlled from the computer? It might be the fist step to a rough path following functionality .
Is it OK if we share the video? It is currently set as unlisted.
No its much more simple than that. I setup the A button to override the throttle and roll input. Throttle goes to about 25% and roll to 1200 degrees, making the left prop go to about 80% and the right prop stop completely. Then it's just a matter of hitting full throttle for a bit, pressing the A button and then letting go at the right moment (requires a little practice).
I rather keep the video unlisted at the moment. It's taken at the office of my work and is not very high quality. I'll try to make a better quality video soon.
Still had no time to work on a smarter flip feature using the sensors. Will attempt that soon.
That combined with a mini state machine (detecing successive "features" in the right order) worked really well. Good luck! Looking forward to seeing what you can come up with!
Using a state machine is a good idea, think I will do the same! Lots of fun, this embedded programming good to compensate for all the web development I do
I haven't flown my crazyflie much recently due to work but I am excited to see this new stuff! Ralph those flips are awesome! Much more stable than trying to do the whole thing manually! Haha.
) That combined with a mini state machine (detecing successive "features" in the right order) worked really well. Good luck! Looking forward to seeing what you can come up with!