Control of an electric wheelchair using multimodal biosignals and machine learning

Publication Name

IEEE/ASME International Conference on Advanced Intelligent Mechatronics, AIM


While the current design of electric wheelchairs is very well established, people with hand disability or motor impairment are limited in their independence. The incorporation of multiple bio-signals in commercial electric wheelchair control systems would increase functionality of the assistive devices for people with restricted or erratic hand movement, consequently increasing the quality of life of such users. This study reports on the implementation of a multimodal bio-signal control system for an electric wheelchair. In this design, wheelchair control via the standard joystick was replaced by a wireless EEG headset with four direction modes - forward, turn (left), turn (right), and neutral/no movement. Control of the movement of the chair is obtained through the analysis of EEG signals, EOG signals and some EMG signals, measured simultaneously. Data is gathered from a minimum number of electrodes, six in total, residing on the low cost commercial Emotiv EPOC+ EEG Headset. Bio-signals are sent from the EPOC+ to the USB dongle in a computer, which filters and processes the data using Python scripts. A moving average gradient thresholding Multiple Classification Ripple Down Rules paradigm has been implemented to ascertain the intended movement direction of the user. The movement intention class is sent via Wi-Fi to a Raspberry Pi with custom PCB which is wired into the joystick plug. The multi-modal bio-signal system proves highly effective, obtaining a classification accuracy of 97.7% over three different trials, though seven instances of false positives were observed, prompting future optimisation of the classification system.

Open Access Status

This publication is not available as open access



First Page


Last Page




Link to publisher version (DOI)