After days of tweaking and callibration, I've finally managed to print this handy Mr Jaws to ensure my coffee stays fresh :-)
The infill is far from perfect... more tuning ahead.
Three days ago, all I had was a lot of wooden pieces, nuts and bolts. Today I have a 3D printer with 100micron precision and this is my first print. There is nothing fancy here, just assembled a Printrbot JR as part of a 3 days workshop organized by Re-Lab. But for me it was a first and a really fun experience mixing hardware, mechanical engineering, electronics and software. Building your own tool and then printing your own parts provides a very satisfying experience.
Now, I'm far from done, there is a lot more tuning to do before I can start printing quality parts, but it is a start and a lot of fun.
I continue my work on attempting to controll the nodecopter using a PID controller. The first results are really encouraging. On this plot, you can see the drone moving forward of 1 meter, from (0,0) to (1,0). The oscillation is due to the speed inertia and you can see the controller successfully bringing the drone to the correct position.
The red line is the position along the x axis. The green line is the drone velocity.
Next step is to fine tune the various parameters to find a balance between the oscillations and convergence speed.
A few weeks ago, I posted my first attempt at estimating the pose of the ARDrone based on integrating the odometry measure. It was a good start but definitively lacked precision.
Thus, the next step has been to implement an Extended Kalman Filter that use a tag detected by the bottom camera (at the landing zone) for correcting the state prediction.
This picture shows the result running the same flight through both a simple estimator and the EKF with correction step. As you can see, in the later, the drone actually lands where it took off. You can also see some subtle corrections as the drone fly by the tag.
All this code is available here and remains quite experimental.
This is my first attempt at estimating the position of an ARDrone by integrating the motion data through time. You can kinda see I have been flying two square patterns.
There is a lot of noise obviously (both sensors and wind) which are the cause of this not so precise tracking. Next step is to integrate observations derived from the camera into a proper Extended Kalman Filter. That's the tricky part :-)
No silos, no paywall, no control by a single entity or government. Pure freedom, at least as pure as emailing or calling someone.
I love it :) When are you joining the conversation ?
"the ISS was in front of the Sun for less than second, so not only did he have one chance at getting this spectacular once-in-a-lifetime shot, but he had only a fraction of a second to snap it!"