基于空间认知的智能导航方法研究

Most navigation systems today achieve their functions with data-driven approaches, which are always designed upon explicit quantitative models and exact mathematical computations, whereas neglect how people acquire, process and reason spatial information. Consequently, intelligent navigation approaches developed upon principles of spatial cognition, which may meet the cognitive habits of human wayfinding, have been becoming research focuses in the fields of navigation. For improving the effectivity and efficiency of navigation systems, several approaches are introduced in this proposal according to the insufficiences of presented navigational methods. The foremost one is a simple and efficient way to automatically extract landmarks from urban environments. On the basis of those landmarks, algorithms of contextual route selection and route directions integrating human spatial cognitive principles and a-priori spatial knowledge are designed, so as to reflect personal cognitive preferences in route selection results, and cause route directions closer to how people communicate about spatial information. The main research topics include: (1) proposing an approach of automatically extracting hierarchical landmarks from points of interest (POI) in urban environments; (2) presenting a complete network model independent, interactive route selection algorithm using hierarchical reinforcement learning (HRL); (3) putting forward a context-adaptive method for the seamless integration of user's a-priori spatial knowledge into turn-by-turn route directions.

由于以数据模型和数学计算为基础的传统导航系统通常忽略用户对空间信息的获取、加工和推理模式，能够使导航过程更加符合人们认知习惯的基于空间认知原理的智能导航方法正在成为导航领域的研究热点。本项目以提高导航系统的有效性、可用性为目标，针对国内外现有研究在地标数据提取、路径规划算法和路径引导方法中存在的不足，提出从城市环境中简捷、高效地自动提取各种地标数据的方法，并在地标数据的基础上实现融入了用户空间认知规律和先验空间知识的自适应路径规划和引导方法，从而使路径规划结果反映不同用户的空间认知偏好，使路径引导过程更加接近人们空间交际的方式。主要研究内容包括：1）提出一种基于城市兴趣点数据的分层地标自动提取方法；2）提出一种不依赖完备路网模型的基于分层强化学习的自适应路径规划方法；3）提出一种面向自然语言且能够实现转向决策和先验知识无缝融合的自适应路径引导方法。

本项目以提高导航系统的有效性、可用性为目标，围绕人们对空间信息进行获取、加工和推理的基本原理和方法，主要从地标提取、路径规划和路径引导三个方面开展理论研究。主要研究内容包括：1）从城市环境中简捷、高效地自动提取多层地标数据的方法；2）不依赖完备路网模型的基于分层强化学习的自适应路径规划方法；3）面向自然语言并无缝融合转向决策和先验知识的自适应路径引导方法。该研究可以使导航系统的路径规划结果反映不同用户的空间认知偏好，并使路径引导过程更加接近人们空间交际的方式。目前，围绕该研究已发表学术论文3篇（包括1篇国际学术会议论文和2篇国内核心期刊论文），参加学术会议和研讨会5次，另有3篇学术论文正处于审稿阶段。其他研究成果正处于实验改进或论文撰写阶段。