The purpose of this thesis is modeling and controlling robot manipulators for pHRI during the visual servoing task. Recent advances in physical human-robot interaction (pHRI) provide the possibility of sharing workspace between humans and robots and coping with tasks involving physical contact with a human under uncertainty, in a stable and safe manner. PHRI can only be accepted when safety and dependability issues are addressed as the central criteria for all the phases of design and control of robots in anthropic environments. Nowadays, robotics is concerned with the emulation of human skills in an artificial context; an obvious requirement for a full interaction with the world is using the vision. In this respect, this thesis explores the problem of using visual information to control the motion of robotic systems, which are equipped with onboard cameras. Since the safest possible solution for the safe interactions is avoiding any undesired contact (collision) with humans or environment obstacles, the obstacle avoidance algorithm for robot manipulator, that executes a visual servoing task, has been addressed. It has been assumed that the environment is observed by depth-measuring sensors which allow measuring the distance between any moving obstacle and the robot. Then, the distances between robot and obstacles are used to generate repulsive vectors, which are used to control the reactive motion of the robot so as to avoid the robot collisions. In order to estimate the velocity of a moving obstacle, the time variations of repulsive vectors has been calculated, which clearly improved the collision avoidance algorithm. The main problem is that the visual servoing task may be stopped during interactions, because the image features may go out of the camera's field of view when the motions for the collision avoidance are being generated. Therefore, the objective of this thesis is to control the motion of a redundant robot manipulator in order to avoid a collision in such a way that the generated motions during obstacle avoidance do not cause the visual features losing. The key idea in this thesis is using the repulsive vectors and modifying them for collision avoidance. This modification has been done by adding rotational motions to repulsive vectors. The proposed approach is verified be several case study on KUKA LWR robot arm. Keywords: Human-robot interaction, Visual servoing, Obstacle avoidance, Redundant robot manipulator.