Due to the complete randomness of sampling, the traditional PRM algorithm was often difficult to be applied to the robot path planning in the working environment, including narrow channels. To this end, an improved probabilistic roadmap method (Improved PRM) integrating global goal-oriented sampling and local node enhancement was proposed and utilized to the path planning of a planar grid map scene and a 6-DOF robot. Firstly, the global goal-directed sampling was combined with the random sampling in the proposed Improved PRM, and the probability of global sampling points falling into narrow channels was raised by the mixed sampling, so as to achieve the heuristic map enhancement. Secondly, nodes in narrow channels were extracted by using the node weight idea, and a local node enhancement strategy based on Gaussian distribution was used to expand new nodes in narrow channels to enhance the connectivity of the map and the success rate of path planning. Finally, the redundant node elimination strategy was presented to optimize the initial path planned by the algorithm. The simulation results of the Improved PRM algorithm in the planar grid map showed that the success rate of the algorithm for robot path planning was more than 89.3%. Besides, the comprehensive evaluation and path quality evaluation were both higher than that of other algorithms. In the simulation experiment of a 6-DOF robot, the average path cost obtained by the Improved PRM algorithm was about 42.7% lower than that of the traditional PRM algorithm. Meanwhile, the probability of successfully passing through the narrow channel was also 68 percentage points higher than that of the traditional PRM algorithm. Therefore, compared with other algorithms, the Improved PRM algorithm had advantages in improving the success rate of path planning, reducing path nodes, and ensuring path quality in the working environment with narrow channels.