Recent advances in brain computer interface (BCI) offers its capabilities to provide a way of communication with non-muscular activity to control computers, robots, and other assistive technologies. This paper exploits the possibility towards a semiautonomous, humanoid robot personal assistant for handicapped subjects. The motor imagery (MI) utilizes the event-related desynchronization (ERD) signal to detect user intentions. The detected user intention is further used to manipulate a small-size humanoid robot. In addition, the robot may avoid collision in terms of a vision-based obstacle avoidance approach so that it can walk in an unstructured environment safely.