Abstract:

To address the issues of slow convergence and excessive path randomness in traditional BI-RRT*, this paper proposed a two-stage optimization framework algorithm, GEP_BIRRT, which combined an improved BI-RRT* algorithm with evolutionary strategies. Firstly, flexible boundary constraints to the BI-RRT* algorithm were introduced to enhance search efficiency by limiting the sampling range, and a metric function to obtain high-quality feasible paths was designed. Second, a path optimization was performed based on a genetic algorithm, in which an optimization region centered on feasible paths was constructed and a multi-objective fitness function was designed to balance path smoothness and safety, ultimately yielding the planned path. Finally, simulation experiments were conducted using MATLAB software. The strong robustness of GEP_BIRRT across three distinct environments was demonstrated by the results. The results show that compared to Informed-RRT* and traditional BI-RRT*, the planning duration is reduced by 59.48% and 20.08% on average, respectively, with average path length reductions of 1.26% and 1.51%, and cumulative turning angle reductions of 32.60% and 40.84%, respectively. It also effectively avoids dynamic obstacles, validating the superiority and feasibility of the GEP_BIRRT algorithm.

Key words: path planning, path extension, genetic algorithm, fast random search