Expressway hedgerow pruning robots need be able to recognize hedgerow and position themselves real-time, to plan an obstacle avoidance trajectory from the starting point to target point based on the position relationship between hedgerows and obstacles. Compared with the traditional industry manipulator, the expressway hedgerow pruning robot manipulator frequently works in unstructured environments with unknown obstacles and irregular scales. It is difficult to establish a mathematical model of obstacles precisely and comprehensively. The problem of real-time obstacle avoidance can be solved by path planning. Thus, aiming at the problem of real-time obstacle avoidance for expressway hedgerows pruning robot manipulator in an unstructured environment, a novel path planning method to avoid obstacle based on perturbed artificial potential field (PAPF) was proposed. According to the distribution of hedgerows and obstacles, simplified models of intelligent pruning robot and sphere enveloping obstacle were established. By considering the geometric relationship between manipulator and obstacle, the collision conditions of manipulator and obstacles were analyzed, and then, the collision avoidance space of manipulator was solved. The traditional artificial potential field method was associated with some problems such as local minimum point (LMP) and goals nonreachable with obstacles nearby(GNRON). In this study, a repulsion field adjustment strategy was presented to optimize the function model of potential field, and a repulsion field perturbation mechanism was introduced to adjust the effect of repulsion in order to flexibly avoid obstacles and successfully reach the target point. The path planning simulation of the designed manipulator was carried out in the collision avoidance space using PAPF. The simulation result shows that the manipulator smoothly jumps out of the LMP and reaches the target point successfully by accurately avoiding obstacles in real time, which verifies the effectiveness and feasibility of the proposed method.