Brain computer interface (BCI) provides promising applications in neuroprosthesis and neurorehabilitation by controlling computers and robotic devices based on the patient's intentions. Here, we have developed a novel BCI platform that controls a personalized social robot using noninvasively acquired brain signals. Scalp electroencephalogram (EEG) signals are collected from a user in real-time during tasks of imaginary movements. The imagined body kinematics are decoded using a regression model to calculate the user-intended velocity. Then, the decoded kinematic information is mapped to control the gestures of a social robot. The platform here may be utilized as a human-robot-interaction framework by combining with neurofeedback mechanisms to enhance the cognitive capability of persons with dementia.