http://nn-os.org/ provides a new computing environment for Neural Network Control applications of all sorts.
Essentially the environment is comprised of two components:
1. Neural Network adaptive learning, currently back-propagation
2. Petri Nets with BSD sockets to orchestrate asynchronous multi-tasking and IO across the networks
For examples see:
The software is optimized using the vectorization flags of the gcc compiler running on BeagleBone board, no special source code manipulations or directives used. Therefore 1000 training for 10x64x10 Neural Network is accomplished in 1 second or less. 30x64x30 networks run in real-time on BeagleBone with minimal memory footprint.
Looking for creative talent interested to apply the code for Arducopter/plane or similar proposals to build novel guidance and maneuver applications.
The code is provided under the Apache 2.0 dual license.
my beaglebone is here.
just need some time to get familiar with this device.
first step will be filter for the gps - easier to start with :-)
i know what the result should be ;-)
Good to hear that, I look for the stuff you need and it is much easier to develop the code more.
Keep in touch
i stuggled a bit to load my favourite linaro system.
for some reasons i also need iced-tea.
beagle support seems not to be a good thing lately when looking at the linaro.
Let me know if we are talking about the same thing:
I use xcode to get most of the compiles done on IMAC, then I actually re-compile on the Beaglebone itself to access the machine dependent devices.
Look into Cloud9.
i'd like to use gdbserver and do the crosscompiling on my linux host.
will check cloud9.
I am about to faint for the night:
http://www.eclipse.org/ use eclipse see if it works.
The cross compiler for the gdb is great, however I am a bit old fashioned and I makefile on the actual host to kick the tires, so to speak.
Please see my makefiles for the CFLAGS to see how to vectorize for the vector processor in Beaglbone, the performance upshut is huge.
The vectorization flags are NOT trivial, so just cut and paste till I get you better ones.
i am using eclipse all day long.
with rse it should play nicely.
alas the a8 is new to me as well as the neon unit.
lot's of options to play with.
I hope to give you NN for the visual pattern matching, I am also hoping to write some SVM code as well:
So you should be able to CLASSIFY in real-time path shapes and simple object shapes and so on.
I would like to use the wireless WIF option for Beaglbone:
A: We experiment training the NN onboard inflight! and see how well the training performs, prior to a serious flight. I hope this to reduce the crash costs.
I am also interested to investigate formation flight, or foraging. For the latter wrote the Petri Net layer for nn-os, so the Beaglebones can be synchronized in their neural network IO.
I do not like CLOUD9 it reminds me of the IDE for the Arduino.
I much rather to open the device drivers for the IP pins, through the open() system call and issue ioctl() with flags on the file-descriptors.
This way you can do fancy IO.
With the code I uploaded for nn-os 0.1 you can map the IO of the NN between several boards and other unix boxes. Therefore you really do not need to directly access the pins, you can actually access the pins from a remote beaglebone with the code in 0.1
This has a performance hit, but for the speeds we are talking about, they are negligible.