New scheme for image space path planning incorporating CAD-based recognition methods for visual servoing

Visual servoing techniques are widely used in the design of autonomous controllers for various robotic arms performing object manipulation in 3D space. Path planning in image space has been studied widely in order to generate the feasible trajectories for the visual servoing controllers. The main idea of image space path planning is to derive a practical paths for image features while taking into account variety of constraints imposed by the image and the physical environment. However, in many practical cases especially in complex and clutter scene spaces extracting the image features is not robust enough and it demands that the target features not to be occluded by the obstacles, robot's body or the target object itself. In this paper, we present a new image space path planning scheme for markerless 3D target that bypasses the feature extraction requirement. Utilizing artificial potential field method the planner generates the desired end-effector path such that a given target remains inside the image boundaries during the robot manipulation task. Experimental results are presented to determine the 2D and 3D generated path with reasonable accuracy for robotic manipulation scenarios.