I. Introduction
Numerous potential users of powered wheelchairs have difficulties to operate these vehicles due, for example, to an inaccurate motor control or to a severe spasticity [1]. Many researchers have worked on this problem since about twenty years following two ways. Firstly we can try to improve the human-machine interface for example by filtering the user's action on the usual control joystick or by implementing new interface sensors. A second way consists in developing “smart wheelchairs”. The “intelligence” of the wheelchair is based on its capacity to perceive its surroundings thanks to various sensors such as ultrasonic telemeters or vision sensors. With these informations the wheelchair can generate autonomous or semi-autonomous motions (obstacle avoidance, wall following…).