Abstract: To solve the problems of strong randomness of sampling points, low utilization rate of sampling points and slow convergence speed of the bidirectional rapidly-exploring random tree algorithm, an improved DEB-RRT (Dynamic Ellipsoid Bidirectional RRT) algorithm is proposed for 2D and 3D environments respectively. On the premise of ensuring probabilistic completeness, the improved algorithm dynamically controls the goal-directedness of random sampling points according to the failure probability of collision detection. After obtaining a feasible path solution, a secondary path optimization process is carried out to make the final path shorter and the corners smoother. The algorithm simulation is implemented on the MATLAB platform, and the results show that compared with other traditional path planning algorithms, the proposed algorithm has shorter path search time, generates a shorter path length and exhibits higher feasibility.