A Performance Study of 14-Channel and 5-Channel EEG Systems for Real-Time Control of Unmanned Aerial Vehicles (UAVs)

Brain-computer interface (BCI), an actively re-searched multi-disciplinary domain, has completely trans-formed the approach to robotic control problems. Researchers have focused on developing algorithms that optimize robotic movement to achieve desired trajectories, and it's a general understanding that route optimization problems are difficult to solve mathematically. Humans, on the other hand, tend to optimize their day-to-day activities intuitively. In order to achieve the desired results, the brain exploits a multi-level filtering approach, where the macro features are weighted in the first layer and the microfeatures in further layers. This optimization inside the brain interestingly, leave distinct traces in electroencephalography (EEG) plots. Based on the observations, we propose to use artificial neural networks to classify the EEG data, which intuitively should give a high classification rate, because the human brain also exploits a network of neurons to classify auditory (time-series) and visual (spatial) data. In this paper, we discuss the performances of 14- channel and 5-channel EEG headsets for robotic applications. Data is acquired from 20 subjects corresponding to four different tasks. Using neural nets, we have been successfully able to classify the EEG input into four different classes. We get an overall classification accuracy of 98.8% for 14-channel and 84.5% 5-channel system. As a real-time demonstration of the interface, the predicted class number is sent to a multi-rotor via a wireless link as an appropriate velocity command.