FuSe: Bio-inspired sensorimotor control for robotic locomotion with neuromorphic architectures using beyond-CMOS materials and devices

在生物学上，哺乳动物的一个关键特性是在复杂环境中高效移动的能力，这是通过称为中央模式生成器(CPG)的神经网络实现的。CPGS使用简单的环境线索为肢体产生有节奏的控制信号模式。CPG发挥作用的一个最好的例子是我们自己绕过障碍的能力。虽然从哺乳动物的角度来看，这样的网络是自然的，但目前还没有有效的方法来使用电子设备和计算机来设计它们。设计这些网络可以彻底改变我们制造未来几代机器人的能力。可以穿越未知和复杂地形的灵活机器人有可能实现商业运输的自主导航，增强洪水和地震期间的灾难应对能力，或者访问发生故障的核电站或太空探索等偏远和不安全的地区。计算机工程硬件方面的进步将有助于这项新技术的创造，这些硬件包括电路、设备和材料，这些都是该项目的核心。有效培训机器人神经形态工程新劳动力所需的技能广度，使得课程设计和与现有框架的整合具有难以置信的挑战性。合作机构之间拟议的NeuRoBots教育联盟将解决这一问题。该联盟的主要目标是合作并实施一项全面的劳动力发展计划，该计划结合了基于证据的最佳实践，以帮助培训新一代工程师和研究人员，他们具备满足半导体行业日益增长的需求的能力。该奖项的目标是利用新兴设备对带有突触和神经元的神经形态网络进行建模、设计和实施，以实现微型机器人的高效和自适应控制。灵感来自于负责敏捷运动的生物神经回路。该项目的技术目标旨在解决三个部分：1)材料跟踪侧重于受物理启发的模型，以了解材料和设备属性，以帮助设计时间动力学；2)设备跟踪开发实现生物启发神经元和突触的新设备和电路；3)系统跟踪将实施灵活的微型机器人，使用CPG网络和强化学习演示生物启发移动。通过混合反馈控制在多个时间尺度上整合非线性时间动力学，并在使用新型设备构建的可扩展节能硬件上实例化CPG网络，目标是展示一个全功能的四足/六足机器人，它可以使用神经科学提供的原理来学习移动。这项工作可以带来神经形态计算、人工智能(AI)、机器人和工业自动化方面的变革性进展，同时为神经调节和自我监督学习的科学提供更深层次的见解。通用神经形态系统的开发模拟了神经科学实验中看到的复杂的神经调节时间动力学，为构建新型计算机提供了途径，以应对大脑倡议和芯片法案中设想的重大挑战，造福于国家和社会。该奖项反映了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。

In biology, a key property of mammals is the ability to move efficiently in complex environments which is enabled through neural networks called Central Pattern Generators (CPG). CPGs produce rhythmic patterns of control signals for the limbs using simple environmental cues. A prime example of CPGs in action is our own ability to navigate around obstacles. While such networks are natural from the perspective of mammals, there is currently no efficient way to engineer them using electronic devices and computers. Engineering these networks can revolutionize our ability to build future generations of robots. Agile robots that can traverse unknown and complex terrains have the potential to enable autonomous navigation for commercial transport, enhance disaster response during floods and earthquakes or to access remote and unsafe areas like malfunctioning nuclear plants or space exploration. The advances in computer engineering hardware including circuits, devices, and materials that form the core of this project will aid the creation of this new technology. The breadth of skillsets that are required to effectively train a new cadre of workforce in neuromorphic engineering for robotics makes curriculum design and integration with existing frameworks incredibly challenging. The proposed NeuRoBots educational consortium among the partnering institutions will address this issue. The main objective of this consortium is to collaborate and implement a comprehensive workforce development plan that incorporates evidence-based best practices to help train a new generation of engineers and researchers, who are equipped to satisfy the growing needs of the semiconductor industry. The goal of this award is to model, design and implement neuromorphic networks with synapses and neurons using emerging devices to achieve efficient and adaptive control in miniature robots. The inspiration comes from biological neural circuitry responsible for agile movement. The technical objectives of this project are designed to address three components: 1) The materials track focuses on physics-inspired models to understand the material and device properties to help engineer the temporal dynamics, 2) the devices track develops new devices and circuits for implementing bio-inspired neurons and synapses, and 3) the systems track will implement agile miniature robots that demonstrate bio-inspired locomotion using CPG networks and reinforcement learning. By incorporating non-linear temporal dynamics at multiple timescales through mixed-feedback control and instantiating CPG networks on scalable energy-efficient hardware built using novel devices, the target is to demonstrate a fully functional quadruped/hexapod robot that can learn to move using principles informed by neuroscience. This work can lead to transformative advances in neuromorphic computing, artificial intelligence (AI), robotics, and industrial automation, while providing deeper insights to the science of neuromodulation and self-supervised learning. Development of general-purpose neuromorphic systems that mimic the complex neuromodulatory temporal dynamics seen in neuroscience experiments offers pathways to build a new class of computing machines that address the grand challenges of the BRAIN Initiative, and advances envisioned in the CHIPS Act, benefiting the nation and society at large.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.