by Vincent de Marignac
A semester project proposed by sinlab at EPFL.
This project makes use of the following open source libraries and pieces of code:
- Emokit by Cody Brocious, Kyle Machulis, Severin Lemaignan and other contributors
- oscpack by Ross Bencina
The goal of this project is to be able to control a quadcopter with your mind. This is done using the Emotiv Epoc EEG headset to get the brainwave data and then controling the ARDrone QuadCopter through an assembled set of existing code and a few ajustments.
To compile the project you will need cmake version 2.6 or higher. Of course the requirements for emokit and oscpack are also regarding this project. So libusb-1.0 and libmcrypt are a necessity if under linux. I do not know about dependecies under any other platform yet.
Please note that this has only been tested on my linux machine, so it might not work correctly on other systems - although I am always trying my best to make it portable.
At the moment the project is far from being finished, I am still having troubles finding a way to send OSC messages (noob I am but noob I shall not remain...). I am currently working with Gael Grosch on the emotiv part. He is helping a lot with the signal processing (actually he's doing it all :). We are experimenting with the openVibe library, trying to figure out what signal corresponds to what state of mind. The idea is to separate waves in terms of frequency, as this is the characteristic that differenciates one state from another. They are all related to certain brain activities like being relaxed, thinking of a simple concept or even moving your limbs. Having done that we will be able to assign states to commands and send them to whichever device we like (the ARDrone in my case).
I have been able to send OSC messages to a program in the Processing language. The problem was actually very simple: the oscP5 library for Processing does not implement MessageBundle yet, so you have to send messages one after the other instead of grouped inside a bundle... Now that I can send and receive messages, the information, unless it is a string, is badly interpreted (probably due to type mistakes). The two OSC libraries (oscpack and oscP5) don't seem to agree on type conventions??
I just figured out the osc reading problem: endianness was set to big endian by default where it should have been little endian. I added the Processing oscARDrone sketch to the example folder.
Added Java as a server for OSC messages. Using Processing inside Java and JavaEEGUtils for computation of FFT as well as plots. Updated battery info from openyou's update.
Added javaml for machine learning, although not well integrated yet. Also started considering P300 event-related potentials. Will probably provide more details on the wiki.