无人驾驶规则的自动提取方法研究
2021-04-09黄治周劲何叶
黄治 周劲 何叶



摘 要:无人驾驶汽车的测试技术主要是通过虚拟测试和场地测试的方法来检测无人驾驶汽车的智能化水平,通过不断的学习和总结经验来提高无人驾驶的安全性, 这种方法耗费大量的人力物力以及时间。为此,从文本的角度出发,利用驾驶行为相关文本,通过NLP技术和基于规则的方法构建一个无人驾驶规则库,依据这个规则库来辅助测试无人驾驶汽车能否满足道路安全要求。实验结果表明,对《中华人民共和国道路交通安全法实施条例》处理后规则提取的正确率为 89.85%,驾考题库文本的正确率为87.33%。
关键词:驾驶规则库;自然语言处理;提取
中图分类号:TP181 文献标识码:A
Research on Automatic Extraction Method of Driverless Rules
HUANG Zhi1 ,ZHOU Jin1 ,HE Ye2
(1.State Grid Hunan Electric Power Limited Company,Changsha, Hunan 410014,China;
2.China Merchants Bank Network Technology Co. LTD,Shenzhen, Guangdong 518057,China)
Abstract:Nowadays, the test technology of driverless vehicle mainly tests the intelligent level of driverless vehicle by means of virtual test and field test, and improves the safety of driverless vehicle by continuous learning and summarizing experience. This method consumes a lot of manpower, material resources and time. Therefore, from the perspective of text, this paper uses the driving behavior related text, through NLP technology and rule-based method to build an driverless rule base, according to which to help test whether the driverless vehicle can meet the requirements of road safety. The experimental results show that the accuracy rate of rule extraction is 89.85% and the accuracy rate of driving test database text is 87.33%.
Key words:driving rule base; natural language processing;extraction
随着人工智能的不断进步和在汽车领域的应用,汽车的智能水平正逐步提高,无人驾驶汽车的测试方法也在不断的更新[1]。无人驾驶汽车作为解决智能交通、安全出行的重要方案,保证其不会造成交通事故并能够避免危险的发生是无人驾驶汽车测试的重要目标。
从已发布的无人驾驶汽车的安全性和可靠性评估中,为了测试事故率极低的目标汽车,仅道路测试就需要行驶不计其数的里程数[2]。而随着虚拟现实技术的发展,虚拟测试逐渐成为主流测试方法之一,对不同驾驶情况进行仿真,针对仿真结果来优化无人驾驶汽车[3]。一些测试研究人员从传统软件测试的角度出发,通过对交通规则进行形式化分析来对无人驾驶车辆行为进行验证。……
