Live Demonstration: Neuromorphic Robot Goalie Controlled by Spiking Neural Network

Abstract

This demonstration shows an implementation of the Robot Goalie using neuromorphic hardware and a Spiking Neural Network (SNN) to control the goalkeeper position. The system consists of four main components: a Dynamic Vision Sensor (DVS128) used as an "eye", a SpiNNaker SpiNN-3 board to run a SNN to predict the final position of the ball and intercept it, an actuator (Futaba S9257) and an Arduino Due microcontroller (MCU). A PCB board was developed to integrate the Arduino and SpiNNaker boards, with power regulators to use a battery pack for the complete setup. The microcontroller acts as the central communication hub which links incoming signals from the DVS, pass them to SpiNNaker, then to receive the signals from SpiNNaker board and generate the instruction to the digital motor where to place the goalkeeper. A simple SNN has been developed to process the visual input and decide where to put the goalie. This solution is different from the classical segmentation and object tracking and is closer to the biological functioning of biological vision. The system represents a real-time and lowpower solution for the task of intercepting incoming objects. This is in particular relevant to autonomous robotic systems which require fast reaction, but at low power consumption.