NCS-FO: Foundations of Biologically Informed Intelligent Machines for Spatial Navigation

虽然哺乳动物进化出了动物界中一些最大的大脑，但尽管果蝇的大脑很小，但它们表现出了非常复杂的能力。它们可以在大多数风力条件下导航和飞行，并在记住刺激位置的同时直接向刺激或避开刺激机动，即使在黑暗中也可以沿任何角度休息或行走。因此，导航敏捷的果蝇较小的大脑为研究时空导航的基本原理提供了一个很好的路线图。最近的研究表明，被称为中央复合体(CX)的大脑区域产生了果蝇的内部指南针和空间记忆。CX还展示了与感官信息的可扩展集成相匹配的简单、高效和健壮的操作。CX电路架构是构建用于空间导航的生物智能机器的首选方案。该项目的首要目标是创造能够将CX空间导航电路的精髓融入神经形态硬件的硅硬件。通过将计算神经科学和计算机工程这两个研究方向交织在一起，该项目具有强大的潜力来开发一种用于空间导航的智能机器原型，以满足自动驾驶汽车的最苛刻要求。该项目的目标有三个：(1)提取CX中导航电路的基础，(2)扩展其计算原理，(3)在硅中实现所产生的功能。要实现这些目标，需要(I)为基于视觉的空间导航开发一个全面的、生物信息的电路体系结构，(Ii)研究编码和存储时空信息的电路机制，以及(Iii)研究不同类型的视觉信息流在塑造CX决策电路中所起的作用。在这种生物架构的启发下，研究人员将开发一种硅CX原型，它将大大增强可编程性、尺寸、功率和准确性，超过传统的神经形态架构。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

While mammals evolved some of the largest brains in the animal kingdom, fruit flies, despite their miniature brains, exhibit remarkably sophisticated capabilities. They can navigate and fly in most wind conditions, and maneuver directly towards or evasively away from stimuli while remembering their location, as well as rest or walk in any angular orientation even in the dark. The smaller brain of the navigationally agile fruit fly thus presents a good road map for studying the fundamental principles of spatio-temporal navigation. Recent studies have shown that a brain area called the central complex (CX) gives rise to the internal compass and spatial memory of the fruit fly. The CX also exhibits ``hardware'' simplicity, efficiency and robust operation matched by scalable integration of sensory information. The CX circuit architecture is a prime candidate for building biologically informed intelligent machines for spatial navigation. The overriding goal of this project is to create silicon hardware that can incorporate the essence of the spatial-navigation circuits of the CX into neuromorphic hardware. By interweaving the two research directions of computational neuroscience and computer engineering, the project has a strong potential to develop an intelligent machine prototype for spatial navigation meeting the most demanding requirements of autonomous vehicles.The goals of this project are three-fold: (1) distill the foundations underlying the navigation circuits in the CX, (2) expand upon its principles of computation and (3) provide an implementation of the resulting functionality in silicon. Achieving these goals calls for (i) developing a comprehensive, biologically informed circuit architecture for vision-based spatial navigation, (ii) investigating the circuit mechanisms that encode and store spatio-temporal information, and (iii) studying the roles that different types of visual information flows play in shaping the decision making circuits of the CX. Informed by this bio-architecture, the investigators will develop a silicon CX prototype that will substantially enhance programmability, size, power, and accuracy over conventional neuromorphic architectures.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.