Parallella Flight Controller / Robot / UAV Daughter Board

[grmccray]

It seems to me that a multipurpose daughter board that contains a good mix of capabilities and was suitable for a wide range of uses is really important.
Clearly, servo outputs and standard sensors such as 3 axis Accels and Gyros, magnetometer, barometer and a GPS connection are necessities.
However, given the extreme capabilities of this board and it's propensity for dealing with image and spatial processing in a highly capable fashion perhaps some consideration could be given to a high performance camera / video / spatial interface: Maybe a Dual frame buffer with direct Multiport FPGA access, possibly configurable as 2D color or 3D point cloud and dynamically binnable.
Seems like it would be best to provide a flexible format so any of a wide variety of input devices (PrimeSense / Leap, TOF Cameras, Laser Scanners, Video Cameras, Stereo laser rangefinders, etc could be used to fill the frame buffer(s).
Just a thought, it does seem like the main thrust of this board is to let your system be able to see and to figure out what to do with what it sees.

[Veqtor]

I'm leaning towards the MPU-9150 9-axis IMU along with the proven MS5611-01BA03 pressure sensor. What about a video input? This I think is a difficult choice... Perhaps waiting for specs on the new capri sensor from primesense would be a good idea... Thoughts?

[grmccray]

Hi Veqtor,
Clearly this board's primary advantage is it's ability to potentially do serious image or point cloud processing, extracting pertinent information for real time adaptive control.
There are lots of devices that can provide this information in assorted formats, Structured light like the assorted PrimeSense devices,
TOF cameras, Stereo Cameras, Laser Scanners and others.
So, it seems to me that providing a suitable input structure that could be used by various of these would be more useful than tying it to a single architecture. (The PrimeSense system suffers mightily in sunlight and it has a fairly narrow and short depth of field.)
Typically, the best format you could get your data in would be a depth image point cloud and if you could assume that is the way it will be provided, then you can make the "video" input suitable to a variety of devices.
And if what you included in the daughter board was a switchable multiple frame, frame buffer that could interface in parallel to your Epiphany chip you could perform very significant spatial and localization information extraction in real time.
The flight controller components, accels, gyros, Magnetometer, baro and GPS are probably better off being remote on a cable anyway.
It's the video and 3D video buffers and input stuff that needs to be intimately connected to your Parallella board.
My plan at this point is to simply hook your Parallella to a PX4.
You don't really need to replace it, you just add, literally, a whole new dimension.

[Veqtor]

Yes I too am leaning towards such a solution, however, this takes away the opportunity to use the epiphany and fpga fabric for other uav related tasks. One more application I could see is creating a kind of convolution of the expected vibration response from various output (you send a signal to a motor or servo and you get a disturbance in your sensors that has nothing to do with the measurement you're trying to make). Then you can convolve this with your output signals and nearly completely nullify erroneous measurement. Possibly simple convolution isn't enough so I'm also considering volterra kernels. This of course needs the ability to perform a lot of multiplications. Furthermore there is the Madgwick algorithm, neural networks, etc.

But yes, being able to offload part of the job, perhaps sensor-fusion incl gps and also saving some i/o for servos/escs by using an external board like the naze32/px4/autoquad could be a good idea. I just hope to be able establish a very fast movement vector-estimation by fusing rgbd optical flow with accels etc

I guess some reason to still consider a self-contained board is:
* Easy of use - someone who isn't into developing and experimenting could download a good software and just snap the parallella into the daughter-board, calibrate and be ready to go. Less parts means FC and sensor algorithms could be updated in one go.
* Less overlap of capabilities - The zynq fpga fabric should be able to replace most of the code in standard FCs without taking up any considerable space.
* Effectiveness - The only part of the parallella that would be occupied in a CV application would be the epiphany and part of the FPGA, wasting the resources provided by the dual core A9, which is in itself overkill for a FC-processor.

And also...
* Promotion - I want to support the parallella and as highly capable device it would be neat to show how it performs a task all by itself.

[ali8]

Let's stay focused.

Guys, you've already mentioned great and really needed components. Looking at what is commonly used by autopilot-community:

1) MPU 6050 [Accelerometer + Gyroscope]
2) HMC5883L-TR [Magnetometer]
3) MPXV5004DP [Abs. Pressure]
4) MPXH6115A6U[Pressure]]
5) MCP9700T-E/TT [[Temperature]

Items 1 & 2 are especially wide-spread and commonly used. For items 3,4 & 5 we can use any equivalents from other vendors, not a big deal, I think. The first two can be combined in a single MPU9050 but this is still new and not tested by the community (see the APM community autopilot).

I believe the above mentioned sensors (in a single daughter board) would be an essential complement to the parallella board.

==============

EDIT: There is a built in temp. sensor in the Zynq. Do we still need a stand alone temp. sensor?

[ali8]

We can adopt an already-available module, which cost only $33 (for +100 quantity) !!

10 DOF IMU: Accelerometer + Gyroscope + Magnetometer + Barometer

http://www.uctronics.com/10dof-imu-ms56 ... p-857.html
also here:
http://www.ebay.com/itm/10DOF-IMU-MS561 ... 1031471551



Or this: http://www.drotek.fr/shop/en/home/62-im ... s5611.html

We only need to adopt this so it can be connected via the PMOD connections available in parallella.

[9600]

[ali8]

Andrew, thanks for the correction. Yes I confused Zedboard with Parallella.

Still, it shouldn't be that difficult. As you see, 10 DOF IMUs are wide-spread (thanks to Arduino community!) so we only need to build something similar which should not be difficult at all (only 3 chips).

And we need to make it such that it is compatible with Parallella connections.

And yes, I believe there is a good demand for such a daughter card.

ali8

[hamster]

[ali8]