摘 要:随着我国机动车数量的不断增加,交通安全隐患问题越来越严重。针对传统模仿学习效率低下的问题,提出一种基于条件模仿学习的辅助驾驶决策模型,在线模仿学习过程中,构建专家经验池和个人经验池来动态分配学习数据,提高辅助驾驶决策的准确度,同时采用图像语义切割和先验知识迁移技术提取图像特征,提高预测的效率和准确性。模拟实验表明,该辅助驾驶决策模型显著降低了平均预测误差,使得辅助驾驶决策更加贴合个人的驾驶习惯。
关键词:辅助驾驶;条件模仿学习;时序语义;图像特征提取
DOI:10.19850/j.cnki.2096-4706.2023.05.018
中图分类号:TP18 文献标识码:A 文章编号:2096-4706(2023)05-0078-04
Research on Assisted Driving Decision-making Model Based on Conditional Imitation Learning
DAI Ruiru
(Department of Applied Technology, Sichuan Preschool Educators College, Mianyang 621000, China)
Abstract: With the increasing number of motor vehicles in China, the problem of traffic hidden dangers is becoming more and more serious. Aiming at the problem of low efficiency of traditional simulation learning, this paper proposes an assisted driving decisionmaking model based on conditional simulation learning. In the process of online simulation learning, expert experience pool and personal experience pool are constructed to dynamically allocate learning data to improve the accuracy of assisted driving decision-making. At the same time, image semantic cutting and prior knowledge transfer technology are used to extract image features to improve the efficiency and accuracy of prediction. The simulation experiment shows that the average prediction error is significantly reduced by the assisted driving decision model, which makes the assisted driving decision more suitable for personal driving habits.
Keywords: assisted driving; conditional imitation learning; temporal semantics; image feature extraction
参考文献:
[1] 高波 . 基于多传感器感知的辅助驾驶技术研究 [D]. 西安:西安工业大学,2021.
[2] 王强 . 基于深度强化学习的自动驾驶控制决策研究 [D].大庆:东北石油大学,2022.
[3] 李沛安 . 基于深度神经网络的交通场景图像语义分割研究[D]. 北京:北京交通大学,2020.
[4] 赵树煊,张洁,汪俊亮,等 . 基于两阶段深度迁移学习的面料疵点检测算法 [J]. 机械工程学报,2021,57(17):86-97.
[5] CODEVILLA F,MÜLLER M,LÓPEZ A,et al. End-to-end driving via conditional imitation learning [J/OL].arXiv:1710.02410v2 [cs. RO].[2022-11-05].https://arxiv.org/abs/1710.02410v2.
[6] 李小艳,宋亚林,乐飞 . 残差密集块的卷积神经网络图像去噪 [J]. 计算机系统应用,2022,31(10):166-174.
[7] 李壮飞,杨风暴,郝岳强 . 一种基于残差网络优化的航拍小目标检测算法 [J]. 国外电子测量技术,2022,41(8):27-33.
[8] 王海军,何艳,周豪,等 . 基于 YOLOv5 网络模型对铁路轨道障碍物检测与识别 [J]. 中国储运,2022(9):86-87.
[9] 林禹,赵泉华,沈昭宇,等 . 改进 SegNet 与迁移学习的遥感建筑物分割方法 [J]. 测绘科学,2022,47(6):78-89.
[10] 王成康,黄李波 . 基于 CARLA 的驾驶仿真平台搭建 [J].佳木斯大学学报:自然科学版,2021,39(3):27-29.
[11] 符莎 . 基于 UNREAL ENGINE 的沉浸式加工中心造型定制设计技术 [J]. 机电产品开发与创新,2022,35(5):162-165.
[12] 朱坚 . 基于深度强化学习的无人车路径规划研究 [D]. 西安:西安工业大学,2021.
作者简介:戴瑞儒(1968.03—),男,汉族,陕西商洛人,讲师,本科,研究方向:电气设备自动控制、工业企业生产过程自动化、电机和仪表检测等。