RDT-RRT: Real-time double-tree rapidly-exploring random tree path planning for autonomous vehicles
Jiaxing Yu,Ci Chen,Aliasghar Arab,Jingang Yi,Xiaofei Pei,Xuexun Guo
DOI: https://doi.org/10.1016/j.eswa.2023.122510
IF: 8.5
2024-04-01
Expert Systems with Applications
Abstract:The complexity of the environment makes rapidly-exploring random tree (RRT) difficult to handle dynamic obstacle avoidance and system constraint in real-time path planning for autonomous vehicles. To handle this issue, this paper proposes a novel real-time double-tree rapidly-exploring random tree (RDT-RRT) algorithm framework. The collision-free path by RRT after B-spline smooth treatment is adopted as the reference path to reduce invalid sampling. Integrating G1 Hermite interpolation with G2 Hermite interpolation reduces the sampling dimension and takes more efficient samples. The optimal distance metric is designed considering dynamic collision detection mechanism and utilized to estimate the costs of the samples in terms of path curvature. Moreover, to have a better understanding of the environment, convolutional neural network (CNN) is embedded to strengthen the collision detection mechanism. By RDT-RRT the smooth, collision-free paths with small curvature changes can be evaluated. For the evaluations of our proposal in global and local planning, the experiments for a real scaled autonomous vehicle are implemented through parallel computing. By comparing with the mainstream RRT-based algorithms, it has shown that in terms of the path quality, our method reduces 92 % and 88 % of cumulative curvature change respectively in obstacle-free and static obstacle scenarios. Compared with other RRT methods, RDT-RRT performs faster convergence rate and Parallel computing increases the updating frequency from 1.1 Hz to 5.5 Hz. The obstacle avoidance capabilities are also improved.
computer science, artificial intelligence,engineering, electrical & electronic,operations research & management science