An Obstacle Avoidance Path Planning Method Based on Improved OPSN for Robotic Arm

Abstract

This paper proposes a three-dimensional obstacle avoidance path planning method for a single-arm manipulator based on an improved Optimization Problem Solving Network (OPSN). To address the difficulties caused by non-convex search spaces, complex obstacle constraints, and the poor performance of conventional swarm intelligence algorithms in narrow feasible regions, the end-effector trajectory is modeled as a polyline with fixed start and goal points and several intermediate waypoints. Path length, trajectory smoothness, and task-related height preference are jointly incorporated into the objective function, while workspace boundary constraints, obstacle safety distance constraints, and minimum height constraints are explicitly embedded into the network structure. In addition, an elite-initialization strategy is introduced to improve the original OPSN, whose initial inputs are purely random and cannot exploit useful historical information across restarts. The proposed strategy maintains exploration in the early stage and generates new initializations from an elite pool in the later stage through adaptive perturbation and weighted combination. Comparative experiments in three representative scenarios show that the improved OPSN achieves superior or competitive overall performance, especially in narrow-passage environments, where it exhibits stronger feasible-solution search capability and shorter planned paths.