A Modular Spiking Neural Network-Based Neuro-Robotic System for Exploring Embodied Intelligence*

Bio-inspired construction of modular biological neural networks (BNNs) is gaining attention due to their innate stable inter-modular signal transmission ability, which is thought to underlying the emergence of biological intelligence. However, the complicated, laborious fabrication of BNNs with structural and functional connectivity of interest in vitro limits the further exploration of embodied intelligence. In this work, we propose a modular spiking neural network (SNN)-based neuro-robotic system by concurrently running SNN modeling and robot simulation. We show that the modeled mSNNs present complex calcium dynamics resembling mBNNs. In particular, spontaneous periodic network-wide bursts were observed in the mSNN, which could be further suppressed partially or completely with global chemical modulation. Moreover, we demonstrate that after complete suppression, intermodular signal transmission can still be evoked reliably via local stimulation. Therefore, the modeled mSNNs could either achieve reliable trans-modular signal transmission or add adjustable false-positive noise signals (spontaneous bursts). By interconnecting the modeled mSNNs with the simulated mobile robot, active obstacle avoidance and target tracking can be achieved. We further show that spontaneous noise impairs robot performance, which indicates the importance of suppressing spontaneous burst activities of modular networks for the reliable execution of robot tasks. The proposed neuro-robotic system embodies spiking neural networks with a mobile robot to interact with the external world, which paves the way for exploring the arising of more complex biological intelligence.