摘要
为了提升无人车在动态复杂环境中的路径规划能力,构建基于分层强化学习的双流时空融合网络,研究多模态输入的高维特征建模与动态适应机制。结合时空概率场与可变形注意力模块,实现对障碍物行为的不确定性建模与滚动优化控制。结果表明,所提方法在仿真环境中具有较低的轨迹偏差、碰撞概率及规划延迟,具有良好的安全性与实时性。
In order to improve the path planning ability of unmanned vehicles in dynamic and complex environments,a dual-stream spatiotemporal fusion network based on hierarchical reinforcement learning is constructed,and the highdimensional feature modeling and dynamic adaptation mechanism of multimodal input are studied.Combined with the spatiotemporal probability field and the deformable attention module,the uncertainty modeling and rolling optimization control of obstacle behavior are realized.The results show that the proposed method has low trajectory deviation,collision probability and planning delay in the simulation environment,and has good safety and real-time performance.
作者
胡季雨
陈奇
HU Jiyu;CHEN Qi(Huaiyin Institute of Technology,Huaian,Jiangsu 223003,China)
出处
《智能物联技术》
2025年第4期73-77,共5页
Technology of Io T& AI
关键词
深度学习
路径规划
双流网络
注意力机制
deep learning
path planning
dual-stream network
attention mechanisms
