Aiming at the problems of RRT algorithm in narrow and long space, including slow convergence speed and rough planned path, the RSS_GN RRT algorithm was proposed. To enhance the algorithm’s convergence speed, a guide node-oriented strategy and a regional sampling strategy was proposed, greatly reducing the search for invalid regions. Next, the sampling angle constraint strategy was introduced to improve the planned path quality, and adopted the method of parent node expansion selection to effectively solve the problem of increased iteration times caused by angle constraint. Furthermore, the algorithm can dynamically reconstruct the map and plan obstacle avoidance path based on the perception information, enhancing its adaptability in the low-speed dynamic environment. The simulation results show that in a narrow and long channel environment, the RSS_GN RRT algorithm reduces path planning time by 77.3%, 51.9%, 84.7%, 98.8%, and 60.3% when compared to the RRT, Goal_bias RRT, RRT under angle constraint, Informed RRT*, and DR-RRT algorithms, respectively. It decreases the number of iterations by 95.9%, 92%, 98.3%, 98.3%, and 89.5% relative to above algorithms. The average curvature of the path is also reduced by 94.1%, 93.2%, 88.7%, 91%, and 92.9%, respectively. The simulation results prove that RSS_GN RRT algorithm’s significant advantages in enhancing planning speed and optimizing path quality. Simultaneously, this paper uses the Ackerman model car to actually measure the local obstacle avoidance ability of the algorithm. After testing, the car can reasonably avoid obstacles that appear during driving.