Addressing the issue of autonomous vehicles’ instability on icy and snowy roads, an improved rapidly-exploring random tree (RRT) path planning algorithm is proposed. Firstly, a dynamic model introducing road adhesion coefficient on icy and snowy roads is established. Secondly, the global target deflection sampling combined with the front pointing and steering angle of the vehicle, combined with the collision avoidance detection and the maximum curvature constraint under the velocity-adhesion coefficient, is used to improve the traditional RRT algorithm problem.Finally, a double quintic polynomial is used for path smoothing to ensure stability, brake constraints, and comfort. The performance of the improved algorithm RRT is compared with that of the traditional algorithm under multi-scenario conditions through the joint simulation of MATLAB-Simulink and CarSim. The experiments show that the improved RRT algorithm significantly improves the path smoothness, reduces the curvature mutation, has short time, high success rate and good stability when driving on ice and snow.