面向人类智慧提取的无人车半物理训练系统

A Training System for Human Intelligence Acquisition for Driverless Car Based on HIL

针对无人车系统控制中复杂程度高,精确描述困难,算法测试成本投入大的问题,为了快速建立这类多目标多因素的控制系统模型并进行快速测试,提出了面向人类驾驶智慧提取的方法。搭建了通过虚拟场景训练来采集人类操控数据并为进一步提炼及测试人类驾驶控制模式提供基础支撑的半物理仿真训练系统平台。其中设计了以电动机为驱动目标,具备测距传感能力的实时控制器原型,建立了考虑复杂交通环境下的虚拟训练场景,完成了虚拟场景、人类操纵及控制器之间的实时通讯。最后通过原型系统的测试,实现了人类操纵数据的提取,为进一步提炼人类驾驶模式建立了数据基础,也证明了该平台在所提方法的应用上是可行的。

A method oriented human driving intelligence acquisition is introduced in this paper.A hardware-in-the-loop train system platform is implemented through using virtual scene system to collect human control data and supply basic support for abstracting and testing human driving control mode.The prototype of real-time controller which aims the driving motor for the target and has ability of measuring distance is designed in this paper.Then,a virtual training system considering complex traffic environment is established,and finish the real-time communication among virtual scene,manipulative behavior controller.Finally,via testing the prototype,realize the abstraction in human driving data.The data base is established for further abstraction of human driving mode.