The brain-computer interface (BCI) technology is a new human-machine interaction technology that realizes people to control external devices directly by thinking (i.e. electroencephalogram, EEG). However, because of the weakness and randomness of EEG signal, it is very complicated and difficult to process and identify the EEG signal recorded by the non-invasive BCI, and the decoding error often occurs. In view of the brain electrical signal decoding error, an experimental paradigm for simultaneous acquisition of spontaneous EEG and evoked EEG was designed, where the subjects generated the error-related potentials (ErrP) based on the decoded feedback of motor imagery EEG. We analyzed the EEG signal two times. The motor imagery EEG, which was the component of the EEG signal, was analyzed at the first time analysis. We classified the motor imagery EEG signal of left and right hand, then analyzed the classification method quantitatively using the Receiver Operating Characteristic (ROC) curves and the area under the curve (AUC). Although the EEG signal were influenced greatly by the individual difference, the AUC values can still reach more than 0.7. Meanwhile, the frequency domain characteristics were analyzed. The activation brain regions of the left-right hand motor imagery are mainly concentrated in the area of the perceptual motor cortex where is responsible for hand motion, but they will also be influenced by the artifacts of the surrounding channels. In the second time analysis, the ErrP was extracted and discussed. Its latency, waveform and amplitude characteristics were studied in the time domain and then a suitable classifier is selected by comparing a variety of classifiers, which classification accuracy is up to 90%. Therefore, the research based on the ErrP signal played a theoretical foundation for applying to the lower limb exoskeleton rehabilitation robot in the future, and ensured the feasibility of applying the command confirmation unit based on ErrP signal to the exoskeleton rehabilitation robot.