P-RT-BFMT: A Prediction-Based Real-Time Bidirectional Fast Marching Tree for Robot Motion Planning in Dynamic Environments

Abstract

Traditional rapidly-exploring random tree (RRT) algorithms and their variants have achieved significant success. However, they still face challenges in the real-time motion planning performance for mobile robots in complex dynamic environments. This article presents an efficient real-time motion planning algorithm for mobile robots, termed P-RT-BFMT, aimed at enhancing motion planning performance in dynamic environments. We develop an efficient real-time planning framework based on the fast marching tree (FMT), which significantly reduces the number of collision checks by uniformly and randomly distributing all samples in the environment during the initial phase, thus outperforming traditional RRT in efficiency. In the planning process, P-RT-BFMT employs the concept of bidirectional tree expansion to efficiently find an initial solution for the global path, further improving motion planning performance in dynamic environments. Additionally, we designed an event-triggered execution scheme based on predicting potential collision risks, enabling dynamic local path replanning. This treatment effectively reduces the frequency of local path replanning seen in traditional algorithms, alleviating computational burdens and execution costs. Extensive comparative experiments are conducted in both simulated and real environments, and the experimental results validate the reliability of P-RT-BFMT in terms of real-time planning performance, robustness, and effectiveness in complex dynamic scenarios.