Aiming at the blindness, randomness of node search, and the discontinuous path planning in the traditional RRT path planning algorithm for unmanned vehicles, a dynamic variable sampling area RRT path planning algorithm is proposed. Firstly, the map information is initialized. The sampling space is divided according to the dynamic variable sampling area formula. Then the sampling area is selected. The initial path planning is completed by using collision detection based on safe distance, probabilistic target offset strategy and multi-level step expansion. Finally, the inverse optimization considering the maximum rotation constraint and cubic B-spline curve is used to fit and optimize the initial path. Simulation results show that compared with the original RRT algorithm, the search time and sampling times of the proposed algorithm in different map environments are reduced by more than 50%, which greatly reduces the blindness and randomness of node search. Compared with other algorithms, the search time of the proposed algorithm is also reduced by more than 30%. The planned path is smooth and meets the constraints of vehicle dynamics.