Abstract: To address the problems of highly random sampling points, low sampling point utilization, and slow convergence of the bidirectional rapidly-exploring random tree (Bi-RRT) algorithm, this paper proposes an improved DEB-RRT (Dynamic Ellipsoid Bidirectional RRT) algorithm applicable to both 2D and 3D environments. While guaranteeing probabilistic completeness, the improved algorithm dynamically adjusts the goal bias of random sampling points based on the collision check failure rate. Once a feasible path is found, a secondary path optimization step is executed to shorten the final path and smooth its sharp corners. Simulations of the algorithm are carried out on the MATLAB platform. The experimental results demonstrate that, compared with conventional path planning algorithms, the proposed method achieves less path searching time, shorter path length, and better overall performance.