RI: Medium: Collaborative Research: A Structure-Math-Function Approach for Designing Robustly Intelligent Synthetic Nervous Systems

机器人正在融入生活的更多领域，不再局限于工厂的可预测环境，执行同样的任务。与人类一起工作的机器人需要更高的智能和适应不可预测环境中不断变化的任务的能力。这项工作通过模拟一种非常聪明、有能力和适应性强的昆虫——螳螂——大脑中的控制系统，开发了一种复杂的机器人控制系统。这项工作有望改变我们对机器人和神经科学领域智能的理解。昆虫相对简单的大脑中的决策模型推动了对更复杂大脑的研究。然后，该模型将用于允许有腿机器人调整其运动，以适应其目标，如协助人类，或其“需求”，如寻找能量或避免危险。这些进步旨在赋予机器人与动物一样的自主权。机器人可以通过训练，不断从经验中学习，即使在新的情况下也能提高效率，而不是针对每种可能的情况进行编程。在市中心一所学校的一个课后机器人项目中，学生们将从创造具有生物灵感设计和神经系统建模独特视角的机器人的实践经验中受益。本研究扩大了螳螂神经系统连续时间动力学模型合成神经系统（SNS）的规模和复杂性，并将其应用于机器人控制。多通道神经记录和刺激技术揭示了昆虫如何通过在整个神经系统中分配计算来简化运动控制。这个项目利用这些技术来理解“较高”层次（处理感觉输入的大脑）的能力是如何直接得到“较低”层次（协调腿部的神经节）的智力支持的。这些数据将用于开发和实施一个SNS来控制六条腿的MantisBot，赋予它在线学习和智能自主。神经生物学将在所有三个具体目标中为这项工作提供信息：1)调查螳螂神经系统的低级智能，并使用结果来提高螳螂机器人低级控制网络的智能；2)研究下行命令与行为之间的相关性，并利用结果开发MantisBot的简化大脑（即高级控制器）；3)研究冲突的视觉输入（例如同时发生的猎物和捕食者）对下行命令的影响，并利用这些发现赋予MantisBot在其整个SNS中分布的鲁棒智能。

Robots are becoming integrated into more areas of life, no longer confined to the predictable environment of a factory, performing the same task. Robots that work among humans require greater intelligence and the ability to adapt to changing tasks in an unpredictable environment. This work develops a sophisticated control system for robotics by modeling the control systems in the brain of a remarkably intelligent, capable, and adaptable insect: the praying mantis. This work promises to transform our understanding of intelligence in both robotics and neuroscience. A model of decision-making in the relatively simple brains of insects advances the study of more complex brains. The model will then be used to allow a legged robot to adapt its movement to suit its goals such as assisting humans, or its "needs" such as seeking energy or avoiding danger. These advances seek to give robots the autonomy that animals have. Instead of being programmed for every possible situation, a robot could be trained, continue to learn from experience, and improve efficiency even in novel situations. At an after-school robotics program at an inner-city school, students will benefit from hands-on experience creating robots with the unique perspective of bio-inspired design and modeling of nervous systems. This work expands the scale and sophistication of a synthetic nervous system (SNS), a continuous time dynamical model of praying mantis nervous system, and applies it to robotic control. Multi-channel neural recording and stimulation techniques are revealing how insects simplify motor control by distributing computation throughout the nervous system. This project leverages these techniques to understand how the capability of the "higher" level (the brain, where sensory input is processed) is directly supported by intelligence in the "lower" level (ganglia that coordinate the legs). These data will be used to develop and implement an SNS to control the six-legged MantisBot, endowing it with online learning and intelligent autonomy. Neurobiology will inform this work in all three specific aims: 1) Investigate the lower-level intelligence of the mantis nervous system and use the results to increase the intelligence of MantisBot's low-level control networks; 2) Investigate the correlation between descending commands and behavior and use the results to develop a simplified brain (i.e. high-level controller) for MantisBot; and 3) Investigate the effect of conflicting visual inputs (e.g. simultaneous prey and predator) on descending commands, and use these findings to endow MantisBot with robust intelligence distributed throughout its SNS.