基于大数据云平台的自动驾驶多场景测试方法
Automatic Driving Multi-scenario Test Method Based on Big Data Cloud Platform
刘彦博 1申赞伟 2郭健美 3孙伟奇 4熊鑫 5黄鹏升6
作者信息
- 1. 上海交通大学电子信息与电气工程学院,上海 200240;华东师范大学数据科学与工程学院,上海 200241
- 2. 上海交通大学电子信息与电气工程学院,上海 200240
- 3. 华东师范大学数据科学与工程学院,上海 200241
- 4. 心动互动娱乐有限公司城北大王寨工作室,上海 200070
- 5. 腾讯科技(深圳)有限公司,广东 深圳 518057
- 6. 北京千挂科技有限公司,北京 100176
- 折叠
摘要
传统自动驾驶测试开发存在协作性差、工具分散等弊端.为此,设计了一种基于云、管、端协同的智能测试系统.在无人驾驶教学平台构建人工智能算法进行数据采集,上传到大数据云平台进行存储、处理和分析;应用深度学习的目标检测框架,对场景下所有目标车辆进行识别;采用车端联合云平台的迭代模式,在车端采集海量道路环境数据,在云端进行模型算法开发和仿真,将算法程序下载到车端.该智能测试系统契合自动驾驶多场景开发模式,提高模型训练速度和开发效率,达到多传感器融合场景的测试标准.
Abstract
Traditional autonomous driving test development has the disadvantages of poor collaboration and scattered tools.For this reason,an intelligent testing system based on cloud,pipe and terminal collaboration is designed.Firstly,an artificial intelligence algorithm is built on the unmanned driving teaching platform for data collection,and then uploaded to the Huawei big data cloud platform for storage,processing and analysis.Secondly,the deep learning target detection framework is applied to identify all target vehicles in the scene;the iterative mode of the vehicle-side joint cloud platform is adopted to collect massive road environment data on the vehicle side,and the model algorithm is developed and simulation verified on the cloud.Finally,the algorithm is downloaded to the vehicle side.This intelligent testing system is in line with the multi-scenario development mode of autonomous driving,improves the model training speed and development efficiency,and meets the test standards of multi-sensor fusion scenarios.
关键词
自动驾驶/云端训练/人工智能开发平台/场景测试Key words
self-driving/cloud training/AI development platform/scenario test引用本文复制引用
出版年
2024
NETL
NSTL
万方数据