CRII: RI: Building A Self-Learning Robot System with Neuromorphic Computing

点击翻译按钮获取中文摘要

Associative memory is a well-known ability in animals that enables efficient learning of relationships among concurrent events. In contrast, current artificial intelligence systems lack a comparable self-learning capability, relying instead on learning from vast amounts of labeled data. That mode of learning is particularly challenging in scenarios where data, power, and other resources are limited. This project will address these challenges by enabling robots to compute and learn in a manner more like human brains, aiming to achieve high energy efficiency and self-learning capability. This research will make the robot smarter, more energy efficient, and able to operate independently in resource-constrained environments. The project will also provide opportunities for college and high school students from the Upper Peninsula of Michigan to participate in interdisciplinary research on neuroscience, robotics, and artificial intelligence.The project addresses critical challenges of data scarcity and energy efficiency in artificial intelligence and robotics through the development of a neuromorphic robot that can learn without relying on labeled datasets and human intervention. First, a neuromorphic robot will be developed that utilizes spiking neural networks, neural assemblies, and various neural coding schemes for perception and navigation. Second, self-learning algorithms will be developed by mimicking the process of associative memory learning. The study will explore signal pathway modification during associative memory to implement associations and memorization. The research has the potential to create new training algorithms that modify the propagation of spiking signals as objective functions. The self-learning capability and energy efficiency of the neuromorphic robotic system will be evaluated in mazes by replicating navigation and learning tasks observed in rodents. This research on self-learning robotics will significantly benefit the development of energy-efficient robotics by reducing their size, weight, and energy budgets while enhancing their independence and intelligence. If successful, the project will introduce a new self-learning method, diverging from data-driven artificial intelligence approaches, to address the challenges of data scarcity and power efficiency.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.