尊重驾驶员意图的共驾型智能车辆主动避障控制方法研究

Considering the deficiencies of driver’s sensory and executive ability under the complex dynamic emergency obstacle conditions, the shared driving intelligent vehicles are taken as the research object. When the driver’s intention is correct, the method of active obstacle avoidance control respect to driver’s intention will be studied, in order to improve the safety and handling stability of shared driving intelligent vehicles. Based on the innovation of theories and key technologies, driver’s random feature will be modelled, the main variables to identify driver’s intention will be found, the obstacle avoidance strategy of optimal vehicle’s states based on approximate discretization technology will be planned, and the shared control method which minimize the intervention of human-machine intelligent controller to driver’s vehicle manipulation will be proposed. The real-time and efficient experimental platform which is taken the driver simulator as the core, will be established for complex road conditions, and the comprehensive evaluation mechanism of active avoidance performance will also be presented. In a variety of typical avoidance conditions, the proposed control methods will be validated and analyzed from multi-aspects. Through the deeply study of this research, the active obstacle avoidance control of shared driving intelligent vehicle with multiple levels of perception and prediction, decision-making and planning, control and execution will be realized. Moreover, it will also improve the development of vehicle’s intelligence.

针对复杂动态紧急障碍工况下驾驶员存在感官和执行能力不足等问题，以共驾型智能车辆为研究对象，深入探讨在识别驾驶意图正确情况下，尊重驾驶员意图的智能车辆主动避障控制方法，提高车辆行驶安全性和操纵稳定性。结合理论方法与关键技术的创新，提出驾驶员避障意图的主表征变量识别与随机特性建模方法、基于离散化近似技术的避障车辆最优状态规划方法、随机预测型共驾避障最优车辆操纵律设计方法，以及表征智能控制器对驾驶员避障操控最小干预的人机共享控制方法。搭建以驾驶员模拟器为核心的可工程化实时、高效实验平台，构建共驾型智能车辆主动避障性能的综合评价体系，在多种典型避障工况下对提出的控制方法进行多角度验证。通过本项目的深入研究，实现具有感知与预测、决策与规划、控制与执行等多层次的共驾型智能车辆主动避障控制，促进汽车智能化技术的发展。

本项目针对避障过程中，不同驾驶员操控能力迥异而导致的潜在安全性问题，提出了在识别驾驶员意图的情况下，尊重驾驶意图的共驾型智能车辆主动避障控制方法。以实车采集的驾驶行为数据作为分析依据，提取出驾驶员脸/眼部信息、方向盘转角熵、车辆状态、环境信息作为建立驾驶意图识别模型的主要特征变量，并在高斯混合隐马尔科夫模型框架下建立驾驶员换道意图识别模型。为考虑在避障过程中动力学横纵耦合的特性，建立横纵耦合的非线性三自由度车辆动力学模型，以最优避障时间为目标，基于高斯伪谱法进行避障轨迹规划。为满足避障过程中的跟踪性和稳定性等多目标，以及动力学和运行学的约束，设计了非线性模型预测控制轨迹跟踪器。针对车辆快速动态系统对预测控制的高实时性、低成本计算和存储等需求，提出车辆避障轨迹跟踪预测控制器的FPGA异构计算加速方案。本项目研究对车辆智能化发展领域中的驾驶员意图识别、轨迹规划、跟踪控制和控制器加速实现等方面起到一定的推进作用，促进车辆智能化水平的发展。