Information

ArduCopter Evolution Team

The ArduCopter Evolution Team is opened to all the pilots who want to explore, develop and test new features added to the official release of the ArduCopter firmware.

Anyone who wants to share freely, with an open mind, some ideas and software improvements for the ArduCopter firmware is welcome here.

JLN

Location: Earth Planet
Members: 150
Latest Activity: yesterday

You will find here some EXPERIMENTAL VERSIONS and MODS under test from the official release of the ArduCopter firmware.


** WARNING these versions are for EXPERIENCED AND BETA TEST PILOTS ONLY **

DISCLAIMER: If you decide to test these codes on a real flying model this is at your own risk and I assume no responsability...


For non explorers, uses the official released version at: http://code.google.com/p/arducopter/downloads/list

Discussion Forum

Suggestions for Arducopter v2.6 (DCM+Quaternion hybrid)

Started by Ozan Aktaş. Last reply by Ozan Aktaş Aug 11, 2012. 31 Replies

hi Arducopter fans there,i will try to give some suggestions regarding arducopter v2.6 before its release. Since there has been some time without any improvement about AHRs algorithms over recent arducopter versions. I think, the best place to begin is DCM algoritm itself.My suggestion is to combine DCM and quaternion methods with their most superior parts. I know there has been a marg quaternion algorithm, but it is unnecessarily complicated. We know that DCM is easy to understand and has a better implemention for drift correcting algorithms and magnetic offset calculations. But DCM algorithm needs to orthogonalize and normalize  rows or colums of the direction cosine matrix, also it is slightly inefficient due to 9 elements of the matrix is to be updated. As for quaternion method, only one vector of element size 4 needs to be updated and normalized.  Actually, it is possible to integrate quaternions while preserving their unity. There is an algorithm for it.  A DCM matrix, which can…Continue

Tags: hard-iron, soft-iron, magnetic, imu, quaternion

BeagleBone: Neural Network Control

Started by Dara Shayda. Last reply by Dara Shayda Jun 23, 2012. 9 Replies

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-propagation2. Petri Nets with BSD sockets to orchestrate asynchronous multi-tasking and IO across the networks For examples see:http://nn-os.org/nnconductor/   http://nn-os.org/petri-net-nnnet2/http://nn-os.org/petri-net-nnnet2-2/ 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…Continue

Book: Neural Network Control of Nonlinear Discrete-Time Systems

Started by Dara Shayda. Last reply by Dara Shayda Apr 16, 2012. 1 Reply

by J. Saranpagani Taylor and Francis GroupAn encyclopedia like list of all known NN applications to control systems all the way from 1940s to now with MATLAB and some C code. All the math explicated with explicit calculations so the coding can be more educated rather than just copying. I will type up all the equations and algorithms, for people who do not have the book.The control systems are from actual machinery used in different industries and all the relevant research papers listed. My hope is to code the known methods in C/C++ and Mathematica (for visualization and tutorials). The tutorials will be in CDF format and animation can be run on a free player. I will port the code into Beaglebone and is possible into Arduino boards. I am coding on IMAC + XCODE as the native platform. I would also publish the error analysis and stats for each method as time goes by. My thoughts are to use the GNU PUBLIC LICENSE 3.0 but please let me know what other licensing is appropirate If any…Continue

Ideas for simulation input to NN

Started by Dara Shayda. Last reply by Dara Shayda Mar 30, 2012. 2 Replies

I got this generous input from one of the XFOIL's usergroup. These ideas give a better simulation for the copter than the one's I have hacked already since the suggested tools are more mature and worked on by the experts in the field. Amazing thing I recognized this week was that the complicated rotor crafts can be off-line trained by NN for complex flying i.e. the customer purchases a copter and downloads the NN offline trained weight matrices for immediate flight e.g. the novice fliers will not destroy thousands of dollars of equipment because they do not have the RC fingers to manually manipulate the copter.From:GEORGE <georgepehli@yahoo.com>Subject: [xfoil] Re: Neural Network ControlDate:March 30, 2012 1:38:24 AM EDTTo:xfoil@yahoogroups.comReply-To:xfoil@yahoogroups.comHello Dara, I think it would be possible to use some of the existing open source and very powerful tools from the wind energy sector for your project. You essentially have 4 rotors that (in a simplified case)…Continue

CAMRAD

Started by Dara Shayda Mar 29, 2012. 0 Replies

For those interested to develop control systems for rotor crafts:http://johnson-aeronautics.com/documents/CAMRADIIoutline.pdf   I was told  that the simulation aspect and calculations for the phase-space can be obtained from this software, rather writing my own which is reduntant. This took might be quite useful for the community here, since slight alterations to the Arducopter is difficult to debug and maintain. dContinue

Offline Training: Take-off in Winds

Started by Dara Shayda. Last reply by Dara Shayda Mar 28, 2012. 1 Reply

HelloFinally got to the actual control algorithm for the Neural Network based guidance system:http://lossofgenerality.devzing.com/blog/2012/03/28/nn-control-takeoff-in-wind/  The test cases were variety of take-offs, straight or bowed or even crashes. However no explicit wind was blow in the offline test cases. Then the quad-copter performed a take-off in a gust, it turned on its NN control when needed and reduced the drift substantially!I added an actual animation at the end of the html file, so please see how the copter wobbles specially at the higher heights. There might be an issue there or simply the errors are too much and cause serious shaking. The conclusion is then clear: The NN control is viable and probably a superior option for small and larger quad-copters, in adverse weather and situations where the CG is compromised or heavy weights are loaded. DaraContinue

Twin Rotor Helicopter

Started by Chris Card. Last reply by Rob_Lefebvre Mar 26, 2012. 1 Reply

Is it possible to configure APM for use with a twin rotor helicopter?A friend of mine was wondering if it could be used with a Chinook type helicopter.He thought that maybe it would take two APM boards, in communication with each other, to perform control.I wonder if modifying the traditional helicopter firmware to re-mix or blend the outputs to two provide the required control signals for the two rotors.I have never looked at how the real Chinook system works, so I can only make guesses.Has this even been tried?Continue

Beta Tests Flights in the field...

Started by Jean-Louis Naudin. Last reply by Dara Shayda Mar 17, 2012. 15 Replies

ArduCopter v2.4.1:beta test flight of LOITER (GPS Position Hold)Here a video of a LOITER test flight of the firmware ArduCopter v2.4.1 (beta) with the latest fix on STAB_D and Throttle.This version tested include the auto-land mod XP2Regards,Jean-LouisContinue

Tags: ARDUCOPTER, BETA

NN accuracy for quad copter

Started by Dara Shayda. Last reply by Dara Shayda Mar 15, 2012. 4 Replies

HellloSingle layer of 64 neurons with 8 input and outputs and learning rate set to 0.1. Training cases created for vertical take off. The Universal Approximate was constructed for smooth function:CONTROL: Vn+1 --> Vn V is {motor1, motor2, motor3, motor4, Roll, Pitch, Yaw, zcoor}Desired from simulator: {motor1, motor2, motor3, motor4, Roll, Pitch, Yaw, zcoor}{0.701, 0.7, 0.7, 0.7, 0.500005, 0.491911, 0.500174, 0.051289}NN calculated thru adaptive learning from a set of 105,000 test cases, offline, zcoor was normalized by a factor maxZ = 30:Obtained from NN learning:{motor1, motor2, motor3, motor4, Roll, Pitch, Yaw, zcoor}{0.751359, 0.69765, 0.743422, 0.710697, 0.572931, 0.580407, 0.620613, 0.107281}  First four numbers are the motors's RPM, normalized between 0 and 1.Complexity:128 multiplications and additions required + 10 multiplications and 4 additions for taylor series expansion of Sigmoid function around 0. 128 word or long word for the learning matrices. The motors' accuracy…Continue

Tutorial: Universal Approximator and NN

Started by Dara Shayda. Last reply by Dara Shayda Mar 6, 2012. 1 Reply

Hellohttp://lossofgenerality.devzing.com/blog/2012/03/06/tutorial-universal-approximator-neural-networks/  Single layer neuron network is trained for simple set of mappings:http://www.wolfram.com/cdf-player/  <---- download this free plugging to see the calculationshttp://lossofgenerality.devzing.com/blog/2012/03/06/tutorial-universal-approximator-neural-networks/ The emphasis was to make the NN training more accessible for programmers without requiring a PhD.The code was actually implemented in Mathematica 8.0 but the pseudo C code is quite accurate. The algorithm was matched to number of papers and 8000 training loop was experimented with. DaraContinue

Comment Wall

Comment

You need to be a member of ArduCopter Evolution Team to add comments!


Developer
Comment by Jean-Louis Naudin on February 22, 2012 at 1:23pm

During the 3rd quarter of 2011, as a student, I have successfully followed the first online Machine Learning course teached by Prof Andrew Ng from the Stanford University. This was a very intense and exciting adventure and I have greatly improved my experience in this field of knowledge. I am very grateful to Prof Andrew Ng and his Machine Learning team from the Standford University.

Jean-Louis Naudin


Developer
Comment by Mark Colwell on February 22, 2012 at 11:51am

Thanks DARPA we accept technology transfer from:

Autonomous Autorotation of an RC Helicopter

Pieter Abbeel1, Adam Coates1, Timothy Hunter2, and Andrew Y. Ng1 1Computer Science Department, Stanford University, Stanford CA 94305

2Electrical Engineering Department, Stanford University, Stanford CA 94305

We thank Garett Oku for piloting and building our helicopter. This work was sup- ported in part by the DARPA Learning Locomotion program under contract number FA8650-05-C-7261. Adam Coates is supported by a Stanford Graduate Fellowship


Developer
Comment by Mark Colwell on February 22, 2012 at 11:23am

I attempted Dr. Ng's Machine Learning course while also getting certification training for MS SharePoint 2010 Admin & Infrastructure design,, way too much overload... So I backed off machine learning, but soon enrolled in Autonomous Vehicle course. I have ST32F4 from Element 14 on its way, It maybe able to handle some of his AI?

Andrews team is amazing, Only a handful of pilots can fly as well as his AI

Testing ACM 2.4 Heli for legacy Raptor with my secret sauce..  

Remzibi OSD with Call_Sign every two minuets,  

Stabilized camera control with operator nudge  (3-way switch controls mix mode to Ch 5 roll & Ch 6 tilt. 

Controlled with TG-9x with FrSky TM module,  & 

3D/DGPS lock for Ublox.

Installed Atolic Deveopment Lite waiting for ST324F4...

Chopping at the bit... while I debug missing  #if  statement.      

 
 
 

© 2014   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service