基于多语义时空域的智能车辆未知区域自主导航方法研究

To overcome the challenges faced by intelligent vehicles or unmanned ground platforms for the autonomous navigation in unknown regions, such as complex and changeable environment, multi-class dynamic target interference and huge differences between different types of platforms, this project makes full use of the advantage of 2D lidar and stereo cameras to build a low-cost, modular and portable environment perception and mapping system. By cooperating with different types of unmanned ground platforms, the system can realize the autonomous navigation in multi-semantic and highly dynamic unknown regions. Specifically, on the basis of multi-semantic targets’ detection-tracking-prediction and the construction of two-dimensional semantic grid map, the search dimension is extended from a single space domain to the xy-t time-space domain. Then, we complete the agent dynamic planning in the time-space domain, and conduct the states decompose for the time-space domain planning result to realize the horizontal control in the time domain and the vertical control in the space domain, respectively. This method solves the problem of random perturbation of dynamic targets, realizes the organic combination of environment perception, positioning, planning and decision-making modules in the real complex dynamic scene. It also provides technical supports for the application of ground unmanned platforms in real natural scenes, such as border patrol, military reconnaissance, building detection, post-disaster rescue, mine maintenance, indoor service and so on to promote the wide application of unmanned platforms in social economy, national defense and military fields.

为克服智能车辆或地面无人平台在未知区域下自主导航过程中面对的环境复杂多变、多类动态目标干扰、平台类型差异巨大等难题，本项目充分利用2D激光雷达和双目相机的各自优势，构建一套低成本、模块化、便于移植的环境感知与地图构建系统，通过配合不同类型地面移动平台，实现多语义、高动态未知区域下的自主导航功能。在多语义目标检测-跟踪-预测及二维栅格语义地图构建的基础上，将搜索维度从单一空间域拓展到xy-t时空域，在时空域下完成智能体动态规划，并通过时空域规划状态解析，完成空间域下的横向控制和时间域下的纵向控制，从原理上解决动态目标随机扰动的问题，实现真实复杂动态场景下感知、定位、规划与决策等模块的有机融合，为智能车辆或地面无人平台在边境巡逻、军事侦察、楼宇探测、灾后救援、矿井维护、室内服务等真实自然场景中的应用提供技术支持，促进无人平台在社会经济和国防军事领域的快速落地。

为克服智能车辆或地面无人平台在未知区域下自主导航过程中面对的环境复杂多变、多类动态目标干扰、平台类型差异巨大等难题，本项目首先完成了视觉、激光传感器融合的环境感知与地图构建系统搭建。然后基于该系统，完成了多语义目标检测-跟踪-预测，并在静动态多要素理解的基础上，完成了时空导航地图构建和时空决策与轨迹规划，分别在虚实平台上完成了算法评测与演示验证。项目开展期间，获国家科技进步一等奖（排4）、中国惯性技术学会优秀博士论文、全国研究生机器人创新设计大赛优秀指导教师等荣誉。入选国家高层次青年人才项目资助计划，入选中国科协青年人才托举工程资助计划，以第一/通讯作者发表论文10余篇，出版学术著作1部并获国家科技著作出版基金和“十四五”国家重点出版物规划项目资助，申请授权发明专利6项。