I saw the montage but how well did it end up working actually? what was the instruction interpretation success rate. My understanding was that these types of eeg bci projects are very limited by the distance from the actual brain activity and interference. I didn't thing this type of project was archivable with such a flawed system
Thats a good question. For signal processing in this project i used a solution from MindAffect. Its called Event Related Potential and it sends a visual stimuli (you can see that in the video as flickering objects) and it reads a response from your brain. Its quite efficient and accurate. However the biggest challenge for me was to get a high electrode signal. I used a soaked sponges with my electrodes and it definitely helps. And you also need to not move to minimize the movement noise. Electrode signal is crucial for calibration process (training a ML model). Its hard to say precisely what is the success rate for intepretation because first you do the calibration and receive the result and then in the testing the model will be predicting which visual stimuli your brain responds to (which robot movement you chose). You can achieve 100% accuracy in your movement prediction but very often I was around 5-6 out of 10 prediction (not calibration process to not get those confused). You can definitely make it more stable by extending the process of calibration (like always the more data in the training the better). Also make sure to not confuse this system with brain motor imagery. BMI is indeed very difficult to achieve good results with noninvasive BCI.
Impressive work! I hope you enjoyed tinkering with our BCI, and the MindAffect team is excited to see your future projects!
Thank you guys :) Im having a lot of fun experimenting with BCI :)
Excellent demo, Pyotr ! Thanks again from the OpenBCI team.
Thank you so much! I'm already thinking about the next projects :)
Nice project!
Love it🎉🎉
I saw the montage but how well did it end up working actually? what was the instruction interpretation success rate. My understanding was that these types of eeg bci projects are very limited by the distance from the actual brain activity and interference. I didn't thing this type of project was archivable with such a flawed system
Thats a good question. For signal processing in this project i used a solution from MindAffect. Its called Event Related Potential and it sends a visual stimuli (you can see that in the video as flickering objects) and it reads a response from your brain. Its quite efficient and accurate. However the biggest challenge for me was to get a high electrode signal. I used a soaked sponges with my electrodes and it definitely helps. And you also need to not move to minimize the movement noise. Electrode signal is crucial for calibration process (training a ML model). Its hard to say precisely what is the success rate for intepretation because first you do the calibration and receive the result and then in the testing the model will be predicting which visual stimuli your brain responds to (which robot movement you chose). You can achieve 100% accuracy in your movement prediction but very often I was around 5-6 out of 10 prediction (not calibration process to not get those confused). You can definitely make it more stable by extending the process of calibration (like always the more data in the training the better). Also make sure to not confuse this system with brain motor imagery. BMI is indeed very difficult to achieve good results with noninvasive BCI.
@@pyotreus thank you!