无人驾驶车辆智能测试评估与环境设计

Intelligence testing under real traffic scenarios remains an important challenge for researchers of autonomous vehicles. There exit a number of problems for current-stage autonomous vehicles testing in China, including: a lack of theory and methodology support for cognitive capability testing, a lack of synthetic and fully quantitative testing methods, a lack of fully quantitative testing method, a lack of standard testing ground that covers typical traffic scenarios, and a lack of real-time testing data collection method..In order to solve these problems, we plan to study intelligence testing methods for autonomous vehicles and establish the standard testing environment and platform in conjunction with the testing methods. More precisely, we will carry out the following three studies:.1. Research on intelligence testing theory and evaluating methods for autonomous vehicles. We will study the relationship between intelligence testing, task-based testing and capability-based testing. Based the obtained relationship model, we can establish the hierarchical task-capability representation framework for autonomous vehicle intelligence testing..2. Research on simulation-based testing and parallel testing for autonomous vehicles, based on the above intelligence testing theory. Integrating the simulation-based testing system and parallel traffic simulation system, we will flexibly and efficiently test autonomous vehicles. Using transportation Big Data, we can improve the performance of testing systems online..3. Research on environmental design for autonomous vehicle testing grounds and building of intelligent integrated testing/validation platform. Using V2X technologies aided by 5G communication, we can collect accurate and real-time testing data and integrate the research achievements from both autonomous vehicles and intelligent transportation systems fields. We will also build a testing ground that contains various typical traffic scenarios for intelligent vehicles.

面向真实交通环境的智能测试是当前无人驾驶车辆研发面临的重要挑战之一，主要表现在无人驾驶车辆环境感知能力的客观评估缺乏相应的理论与方法支撑，测试量化程度不高、测试场地未包含典型交通场景、测试数据收集能力不足等。.本项目将重点开展无人驾驶车辆智能测试评估方法研究，构建与之适应的标准测试环境与平台。主要研究内容有：1）无人驾驶车辆的智能测试理论与评估方法，探索智能评价、任务评估和能力评测的联系，建立无人驾驶车辆智能的分层任务-能力表达体系；2）基于智能测试理论的无人驾驶车辆仿真测试与平行测试系统研发，通过仿真测试系统与人工交通系统的互动开展集成测试，借助交通大数据，在线改进测试系统的性能；3）面向无人驾驶车辆现场测试的环境设计技术研究和智能综合测试验证平台建设，基于5G通信的V2X技术实现测试数据的实时采集，将无人驾驶车辆技术与智能交通系统研究成果有机融合，建设一个含各类典型交通场景的测试基地。

面向真实交通环境的智能测试是当前无人驾驶车辆研发面临的重要挑战之一。为了解决相关问题，本项目进行了下述三个方面的研究：.1. 提出无人驾驶车辆的智能测试理论和评估方法，研制无人驾驶车辆单项与综合智能的测试标准、评估方法及与之配套的环境设计技术。.2. 研发实现无人驾驶车辆的仿真测试和平行测试系统，实现了无人驾驶车辆的单项和综合仿真测试，以及对于场地测试车辆的实时打分。.3. 建设无人驾驶车辆的现场测试基地和智能综合测试验证平台，基于V2X技术实现无人驾驶车辆测试数据的实时采集，降低测试成本，提高测评的科学性和公平性。.经过两年的努力，实现了申请书中提出的各项研究目标，发表多篇高水平学术论文（其中SCI索引论文9篇），获得多项专利和软件著作权，提供了满足今后5年无人驾驶车辆智能测试、比赛并与真实城区智能交通环境相融合的场地，场地面积大于2平方公里且含各类典型交通场景。协助成功举办了2016年和2017年的中国智能车未来挑战赛，首次在国内举办了高速公路测试比赛和虚拟离线测试比赛。