Cognitive graphs: The geometry of spatial knowledge

适应新环境，发现并记住资源的位置是人类生存的关键技能。在探索一个新的城市时，游客经常在街道上漫步，目的是参观个别景点，欣赏特定的景色。通过这种流浪，他们了解了城市街道的网络。当他们回到酒店时，他们不太可能走回头路。他们也不会走一条直线捷径。相反，他们会根据对周围环境的了解来确定一条利用城市街道回家的路径。游客凭经验怎么能找到回去的路，却找不到一条直线捷径呢？这个项目解决了人类导航研究中的一个具有挑战性的问题：我们的导航知识的潜在几何结构是什么？它真的像人们普遍认为的那样是一张认知地图吗？是更粗糙更流畅的东西吗？我们建议底层结构是一个图？网络中各个地方之间的一系列连接，很像城市街道。本提案的总体目标是使用行走虚拟现实（VR）方法测试关于图知识及其属性的假设。这项研究的结果有可能影响其他领域，包括机器人、神经科学、数学和地理。本文提出的跨学科研究将建立对空间取向理论和认知机制的深刻理解，具有深远的影响。对人类导航系统基本特性的更多了解将导致改进和更有效的电子导航和GPS系统、自动驾驶汽车、应急响应培训和交通标志。chratil博士致力于培养下一代科学家，并增加STEM领域中代表性不足群体的参与。作为该提案的一部分，christl博士将开发一个关于虚拟现实研究方法的研讨会，以培训这项令人兴奋的技术的研究人员。她作为教师和导师积极参与鼓励妇女和女孩参与科学的项目。这个项目特别测试了一个假设，即最有可能的导航知识结构是一个标记图，它包含了局部度量信息，但在整个环境中不是全局一致的。我们之前对认知图的概念进行了初步测试，表明图形空间知识被优先使用，而不是简单地学习地点之间的路线（christtil & Warren, 2014）。然而，目前还没有对标注图知识进行系统的测试。在学习过程中，图知识是如何构建的，或者标记图是否适用于所有的个体和情况，这些都知之甚少。在这里，我们提出了一个框架，详细说明了图知识的理论贡献者。我们还将测试图知识和其他空间知识分类之间的关系。在一系列实验中，我们将使用完全沉浸式虚拟现实来测试关于空间知识本质的假设。首先，我们将通过将标记图知识与其他类型的空间知识进行比较，并测试标记图知识的初步预测，来表征认知图的本质。接下来，我们将测试标记图知识的约束，例如跨上下文的泛化性、与其他分类法的关系以及空间偏好。最后，我们将研究学习标记图知识的机制，特别是注意力和主动决策的部署。这项研究通过对比局部和全局的空间理解水平，并将标记图与其他空间系统进行对比，从而跨越了分析的层次。这项工作的结果将为人类空间知识的基本形成提供关键的见解，并将为机器人和神经科学领域做出贡献。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Being oriented in a new environment and finding and remembering the locations of resources are critical skills for human existence. When exploring a new city, tourists often wander through the streets, with the goal of visiting individual sites and taking in specific views. Through this wandering, they gain an understanding of the network of the city streets. When returning to their hotel, they are unlikely to retrace their steps. They also will not take a straight-line shortcut. Instead, they will draw on what they have learned about their environment to determine a path home using the city streets. How are tourists able to find their way back, based on experience alone, yet unable to determine a straight-line shortcut? This project addresses the challenging problem in the study of human navigation: What is the underlying geometric structure of our navigational knowledge? Is it truly a cognitive map, as is commonly assumed? Is it something coarser and more fluid? We propose that the underlying structure is a graph?a series of connections between places in a network, much like city streets. The overarching goal of this proposal is to test hypotheses about graph knowledge and its properties using walking virtual reality (VR) methods. Outcomes of this research have the potential to impact other fields, including robotics, neuroscience, mathematics, and geography. The interdisciplinary research proposed here will establish a deep understanding of the theory and cognitive mechanisms of spatial orientation, with far-reaching impacts. Greater knowledge of the basic properties of human navigational systems will lead to improved and more effective electronic navigation and GPS systems, self-driving vehicles, emergency response training, and transportation signage. Dr. Chrastil is committed to training the next generation of scientists and increasing the participation of underrepresented groups in STEM fields. As part of this proposal, Dr. Chrastil will develop a workshop on research methods for virtual reality to train researchers on this exciting technique. She actively participates as a teacher and mentor for programs that encourage participation of women and girls in science.This project specifically tests the hypothesis that the most likely structure of navigational knowledge is a labeled graph, which incorporates local metric information but is not globally consistent across the environment. We have previously conducted an initial test of the idea of a cognitive graph, demonstrating that graph spatial knowledge is used preferentially rather than simply learning routes between locations (Chrastil & Warren, 2014). However, a systematic test of labeled graph knowledge has not been conducted. Little is known about how graph knowledge is constructed during learning or whether labeled graphs are used across all individuals and situations. Here, we propose a framework detailing the theoretical contributors to graph knowledge. We will also test the relationship between graph knowledge and other spatial knowledge taxonomies. In a series of experiments, we will use fully-immersive virtual reality to test hypotheses regarding the nature of spatial knowledge. First, we will characterize the nature of cognitive graphs by pitting labeled graph knowledge against other types of spatial knowledge and testing a primary prediction of labeled graph knowledge. Next, we will test the constraints on labeled graph knowledge, such as generalizability across contexts, its relationship to other taxonomies, and spatial preference. Finally, we will examine the mechanisms of learning labeled graph knowledge, in particular the deployment of attention and active decision making. This research cuts across levels of analysis by contrasting local and global levels of spatial understanding and by pitting labeled graphs against other spatial systems. The results of the proposed work will provide critical insight into the fundamental formation of human spatial knowledge and will contribute to the fields of robotics and neuroscience.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.