This paper presents our contribution to vision based robotic assistance for people with disabilities. The rehabilitative robotic arms currently available on the market are directly controlled by adaptive devices, which lead to increasing strain on the user's disability. To reduce the need for user's actions, we propose here several vision-based solutions to automatize the grasping of unknown objects. Neither appearance data bases nor object models are considered. All the needed information is computed on line. This paper focuses on the positioning of the camera and the gripper approach. For each of those two steps, two alternative solutions are provided. All the methods have been tested and validated on robotics cells. Some have already been integrated into our mobile robot SAM.