Rapidly-exploring random tree (RRT * ) algorithm has been widely applied to path planning problem for quadrotor, which takes a great amount of static and dynamic constraints into account. However, conventional RRT * algorithm is suffering low convergence rate and efficiency in below-canopy environment, where is usually occupied with narrow aisles and uneven distributed obstacles. In order to enrich the forest information database and obtain information efficiently in blow-canopy, an improved variable-step RRT * algorithm has been proposed in this paper. In the proposed algorithm, the sampling nodes are determined by the target bias sampling strategy with variable probability. To further increase the efficiency of node generation and balance the searching time and security in areas with different obstacles densities, the eventtriggered step size extension has been proposed according to hyperbolic tangent function. After that, both Euclidean distance and angle constraints are adopted in the cost function of node connection optimization. Then, the path is further optimized by path trimming and Bezier curves smooth method. Finally, the demonstration of the proposed algorithm is presented to show its efficiency, and the comparison result with existing RRT-like algorithms indicates that the performance of our algorithm is better than RRT, RRT * and DDRRT in terms of convergence and accuracy. Comparing to other classical path planning methods such as A * , Artificial potential field and Fuzzy-logic, our method also shows its superiority.INDEX TERMS RRT * , Path planning, quadrotor, below-canopy environment, variable-step.