文献类型：期刊文献

中文题名：空间智能:驾驶态势中的活地图

英文题名：Spatial intelligence:a living map in driving situation

作者：李德毅[1];王留召[2];杜煜[3];殷嘉伦[4]

第一作者：李德毅

机构：[1]军事科学院系统工程研究所,北京100091;[2]中国测绘科学研究院摄影测量与遥感研究所,北京100036;[3]北京联合大学机器人学院,北京100101;[4]清华大学车辆与运载学院,北京100084

第一机构：军事科学院系统工程研究所,北京100091

年份：2026

卷号：21

期号：2

起止页码：498-509

中文期刊名：智能系统学报

外文期刊名：CAAI Transactions on Intelligent Systems

收录：;北大核心:【北大核心2023】；

语种：中文

中文关键词：空间智能;活地图;路权;驾驶态势图;随车而动;变粒度栅格;对数极坐标;贝叶斯滤波;分层记忆;自动驾驶

外文关键词：spatial intelligence;living map;right of way;driving situation map;moving with the vehicle;variable-resolution grid;logarithmic polar coordinates;Bayesian filtering;hierarchical memory;autonomous driving

摘要：本文研究随车而动的驾驶员眼中的活地图在无人驾驶过程中的实时生成技术,即驾驶空间智能。在线生成并持续更新车辆当前路权的地理空间表征,乃是机器驾驶脑中驾驶决策的基础数据,对无人驾驶车辆的认知工程实现具有基础性意义。以变粒度对数极坐标栅格为统一形式化语言,给出活地图系统生成框架;能从传统意义的静态地图问题提升为随车而动的地理空间的状态估计问题,并被持续更新与估计;能承载多源异构的道路传感器感知与先验知识的融合,以贝叶斯滤波为核心机制维护其后验分布;用瞬时记忆—工作记忆—长时记忆维持地图的连续性,并用实时规划实现驾驶活地图的认知闭环。如果路上没有任何其他移动障碍物,活地图就可以让车在不同路段以不同速度从出发地自主驾驶到达目的地。
This paper examines the real-time generation technology of the live map from the driver’s perspective,who follows the car during the process of autonomous driving,known as driving space intelligence.Online generation and continuous updating of the geospatial representation of the vehicle’s current road right is the basic data of driving decision-making in the machine driver’s mind,which is of fundamental significance to the realization of cognitive engineering of unmanned vehicles.Taking a variable granularity logarithmic polar grid as a unified formal language,the generation framework of the active map system is given.It can be upgraded from the traditional static map problem to the state estimation problem of the geographical space that moves with the car,and it can be continuously updated and estimated;It can carry the fusion of multi-source heterogeneous road sensor perception and prior knowledge,and maintain posterior distribution with Bayesian filtering as the core mechanism;Instantaneous memory-working memory-longterm memory is used to maintain the continuity of the map,and real-time planning is used to realize the cognitive closed loop of the driving map.If there are no other moving obstacles on the road,the mobile map can enable the car to drive from the starting point to the destination at varying speeds on different sections.