面向真实道路驾驶行为训练环境研究

Research on Driving Behavior Training Environment for Real Road

为方便驾驶行为智能体的训练与算法研究,运用强化学习算法构建面向真实道路行车规则的仿真环境。提出基于高德开放平台的路径规划及交通状况、兴趣点搜索、天气查询、坐标转换应用程序编程接口,部署人工智能平台OpenAI gym仿真接口环境UBI_Robot-v0的设计方法,有助于在真实道路仿真环境下对驾驶行为智能体进行训练的研究。测试实验结果表明,通过对驾车的环境状态和动作进行设计,实现仿真车主从驾车环境交互数据中多风险维度结合下学习最优的驾驶行为,展现出良好的驾驶行为训练及研究前景。

In order to facilitate the training and algorithm research of intelligent driving behavior, a simulation environment oriented to real road driving rules is constructed based on reinforcement learning algorithm.The application programming interface of route planning and traffic conditions, search for interests, weather query, and coordinate conversion based on AutoNavi open platform is proposed, and the design method of UBI_Robot-v0,the simulation interface environment of artificial intelligence platform OpenAI gym is deployed, which is helpful for research on traning driving behavior agents in a real road simulation environment.The experimental test results show that by designing the environmental state and action of driving, the owner of simulated car can learn the optimal driving behavior from the combination of multiple risk dimensions in the interactive data of the driving environment, showing a good prospect for driving behavior training and research.