Dynamic RRT: Fast Feasible Path Planning in Randomly Distributed Obstacle Environments

For path planning problems based on Rapidly exploring Random Trees (RRT), most new nodes merely explore the environment unless they are sampled directly from the subset that can optimize the path. This paper proposes the Dynamic RRT algorithm, which aims to plan a feasible path while balancing the convergence time and path length in an environment with randomly distributed obstacles. It estimates the length of a path from the start node to the goal node that is constrained to pass through an extended tree node, and this path length is heuristically taken as the major axis diameter of the informed subset. Then new node sampling is performed directly in this subset to optimize the estimated path. In addition, the idea of dynamic programming is employed to decompose the planning problem into subproblems by updating the node selected through Pareto dominance as the new start node to optimize the distance to the goal. Simulation results confirm the performance of the proposed algorithm in balancing the convergence time and path length and demonstrate that the convergence time is faster than that of RRT, while the path length is better than that of RRT*. Dynamic RRT also shows better performance than Lower Bound Tree-RRT(LBT-RRT), and Informed RRT* takes more time to compute a path of the same length.