We present an experiment we carried out to determine the influence of the video feedback on the brain-computer interface performance of a system we designed to steer a humanoid robot. The interface is based on the well-known steady-state visually evoked potentials and the stimuli are integrated into the live feedback from the robot's embedded camera. Five users controlled the HRP-2 humanoid in an experiment designed to measure the performance of the intentions' recognition system. A novel approach in the training phase is also experimented to understand and compensate performance loss due to the dynamic nature of the video feedback of the robot during walking motions. It results that this feedback induces a performance loss; we propose an effective solution to overcome this problem. The detailed results of these experiments are reported in this paper and we discuss the possible causes of performance loss under such conditions.