Aiming at the problems of uneven distribution of sampling points, redundant sampling points, complex path map construction and overmuch path folds when using traditional probabilistic roadmap (PRM) algorithm for aircraft wing box assembly robots, a path planning method based on improved PRM algorithm is proposed. Firstly, the Halton sequence is adopted to optimize the sampling strategy to ensure the uniform distribution of sampling points in the configuration space, so as to improve the sampling quality. Secondly, an optimization strategy for redundant points in elliptic region based on the control points is designed, and the locality sensitive hashing (LSH) function is introduced to reduce redundant sampling points in the configuration space and to optimize the construction and searching efficiency of the probabilistic roadmap. Finally, a B–spline curve is used for smoothing the planning path to meet the actual motion constraints of the wing box assembly robot. The simulation experimental results in 2D and 3D demonstrate that the improved PRM algorithm reduces the planning time by 41.1% on average in 2D space and by 68.43% on average in the high-dimensional configuration space of the robotic arm, compared with the traditional PRM algorithm. Meanwhile, the generated paths are more optimized, which significantly improves working efficiency of the wing box assembly robot.