Vision-Based Potential Field Path Planning for Robot Obstacle Avoidance Under Field-of-View Constraints

Path planning improves the performance and robustness of vision-based robot control in unstructured environments. Visual servo path planning usually focuses only on the path of the camera in the Cartesian space or the path of the feature points in the image space, which means that the collision between robots and obstacles in the workspace cannot be avoided. This paper proposes a vision-based potential field path planning method for robot manipulators with field-of-view constraints in the image space and obstacle avoidance in the Cartesian space. A hybrid potential field function is proposed to integrate Fo V limits in the image space and obstacle avoidance in the Cartesian space. A probability-based search strategy is presented for path planning to avoid oscillation caused by gradient descent. The proposed method is applied to a collaborative robot with 7 degrees of freedom equipped with an eye-in-hand camera. Experiments in two different environments have verified the superiority of the proposed method regarding spanned image areas, camera path lengths, iteration times, and obstacle avoidance compared to the classical gradient descent-based method.