非结构环境下工程车辆自主作业关键技术研究

The intellectualization of construction vehicles has generated considerable recent research interest for its pivotal role in improving the operation efficiency and guaranteeing the safety of the operators. Nevertheless, in the account of the factors such as the complex working conditions, the various objective materials and the involvement of complicated phasing arrangement, it is difficult to meet the requirements of safety, accuracy, and reliability for the current level of autonomous construction. And this technical limitation restricted the intellectualization process of construction vehicles. In order to break through the bottleneck, this research focuses on the autonomous operation of construction vehicles in unstructured environments based on techniques of multi-information fusion and machine learning. Technically, the research method combining working mechanism study, simulation, and field experiment together will be taken to achieve the research object. To be more specific, the first step is to identify the unstructured environment including information of operational objects, moving obstacles, preceding drivable area, collaborative vehicles and etc. Secondly, the image of obstacles should be proceeded by expansion based on their own characteristics, which serves the hybrid-algorithm-driven planning of the optimal anti-collision route for safe driving. Then, the fuzzy reasoning rules would be applied in developing the adaptive cruise control system to track the route both precisely and efficiently. Finally, after arriving at the job site, the strategies for the autonomous operation would be generated depending on the working conditions, guided by which, the specific tasks would be accomplished. The achievements of this research will provide new research foundations and methods for intellectualization of construction vehicles, which will be of great theoretical and practical significance in promoting our national research in relative subjects such as vehicle information acquisition, environmental identification, path planning, collaborative operation and etc.

工程车辆智能化是提高工程施工效率、保障作业人员生命安全的有效手段，已成为相关领域的研究热点。由于工程车辆工作环境复杂、作业对象不确定性高、作业流程繁复等特点，使得其自主作业的安全性、高效性和可靠性等难以达到要求，制约了其智能化进程。为突破该瓶颈，本项目采用机理－仿真－实验相结合的方法，基于多信息融合与机器学习开展非结构环境下工程车辆自主作业研究。首先，对包含作业对象、动态障碍物、可行驶区域、协同工作伙伴等在内的非结构环境进行识别；在此基础上，综合人工势场与蚁群算法的优点，形成动态路径规划算法并搜索最优安全行驶路径，利用模糊推理规则开发其自适应行走控制系统以完成对轨迹的精准、高效跟踪；最后，设计兼容性强、可扩展性高且接口丰富的作业实施系统架构并根据具体任务制定作业策略，完成自主作业。本课题的研究成果将为工程车辆智能化提供新的研究基础和方法，具有重要的理论和现实意义。

开发安全、高效、可靠的自主作业技术是工程车辆向智能化发展中不可回避的基础性科学问题，同时是行业发展中迫切需要解决的关键技术问题。本项目针对传统算法在非结构环境下进行点云配准时受噪声干扰大、配准精度差、易受两点云初始位置影响等问题，提出了一种基于点云曲率特征相似度的改进算法。在应用主成分分析法进行粗配准的前提下，对三维点云进行分割以加快近邻点搜索速度，基于离散指数映射和模板进行采样以简化点云数据并提高配准效率，基于曲率特征相似度搜寻欧氏距离最近点以提高配准精度；非结构环境现场实验验证了方法的准确性和可行性。提出了对铰接工程车辆前后桥进行同步路径规划和跟踪控制的方法。在转向半径分析的基础上建立了运动学模型及预瞄跟踪误差模型，并仿真验证了运动学模型的准确性；在规划出前桥“V”形工作路径的同时根据运动学模型规划出后桥同步路径；基于预瞄跟踪误差和强化学习得到了控制量，实现了前后桥的同步路径规划和跟踪控制，并利用联合仿真验证了规划和控制算法的准确性；模型实验显示本项目提出的方法在精度和稳定性等方面均具有良好表现。针对自主挖掘作业对轨迹的实时性、连续性需求，提出了任务驱动的工作装置连续轨迹规划方法。建立了单次轨迹规划模型，并基于PSO对工作装置各关节的角度曲线进行多目标优化；建立了PINN模型，并以PSO优化所得的结果作为训练样本、以挖掘点信息为输入、以轨迹参数为输出进行模型训练；将PINN模型嵌入到连续轨迹规划框架当中，针对典型任务进行连续轨迹规划。实验结果表明，本项目所提方法在挖掘时间、满斗率和能耗三方面的综合性能达到了最优。依托本项目发表学术论文14篇，所有学术论文均被EI收录，其中11篇被SCI收录（他引61次），申报发明专利8项（3项获授权），培养4名博士研究生和9名硕士研究生，培养5名青年教师。本项目提出的非结构环境识别技术、轨迹规划和跟踪技术以及作业决策技术，为工程车辆的无人化和智能化提供了基础。