Development of a Motor Imagery Based Brain-computer Interface for Humanoid Robot Control Applications

This paper focuses on the developments of asynchronous motor imagery (MI) based brain-computer interfaces (BCIs) applications, signal processing and machine learning to provide some basic capabilities for consumer grade products. For the proposed MI detection technique, two channels of FC5 and FC6 according to 10-20 system over primary motor area are used to recognize 3 mental tasks of tongue, left hand and right hand movements. The amplitude features of EEG signals are extracted from power spectral analysis especially in mu rhythm (8 - 12 Hz) and low beta wave (12 - 16 Hz) bands. MI features were obtained from offline analysis, and then applied to neural network (NN) with particle swarm optimization (PSO). The classification paradigm then applied to real-time BCI for humanoid robot control applications in terms of recognized MI classes from subjects. According to the experiments of 45 trials for a healthy subject, the NN-based MI recognition accuracy with PSO is 91%.