This paper presents a self-adaptive general type-2 fuzzy inference system (GT2 FIS) for online motor imagery (MI) decoding to build a brain-machine interface (BMI) and navigate a bi-pedal humanoid robot in a real experiment, using EEG brain recordings only. GT2 FISs are applied to BMI for the first time in this study. We also account for several constraints commonly associated with BMI in real practice: 1) maximum number of
electroencephalography (EEG) channels is limited and fixed, 2) no possibility of performing repeated user training sessions, and 3) desirable use of unsupervised and low complexity features extraction methods. The novel learning method presented in this paper consists of a self-adaptive GT2 FIS that can both incrementally update its parameters and evolve (a.k.a. self-adapt) its structure via creation, fusion and scaling of the fuzzy system rules in an online BMI experiment with a real robot. The structure
identification is based on an online GT2 Gath-Geva algorithm where every MI decoding class can be represented by multiple fuzzy rules (models). The effectiveness of the proposed method is demonstrated in a detailed BMI experiment where 15 untrained users were able to accurately interface with a humanoid robot, in a single thirty-minute experiment, using signals from six EEG electrodes only.