Neuromorphic Robotics on Conventional Low-Power CPU Hardware

Abstract

This work presents an all-in-one neuromorphic system that emulates neuromorphic computing using a conventional low-power CPU. The system runs a Virtual Machine on a Raspberry Pi 5, integrating Python and specific packages for Spiking Neural Networks (SNNs) simulation in real time. It serves as a cost-effective alternative to specialised neuromorphic hardware, leveraging the Raspberry Pi’s built-in GPIO and USB ports for sensor interfacing. A robotic goalkeeper platform demonstrates the system’s viability, achieving an accuracy of 84% in various conditions with a maximum power consumption of 20 W. This approach offers a competitive solution for low-latency, low-power robotic applications, improving upon previous results in both online and offline setups. One of the key advantages of our solution is that it offers an affordable platform for developing neuromorphic robotic systems, especially when dedicated neuromorphic computing devices and interfacing boards are unavailable or too costly.