Neurally-inspired robotic controllers implemented on neuromorphic hardware

Abstract

Neurorobotics aims to use models of biological neural systems to design efficient and adaptive robotic controllers.For complex, perception-driven behaviors, such neuronal controllers require substantial computational time and resources when simulated in software.However, these controllers can be implemented in an efficient way using neuromorphic hardware systems.Here, we present two neural-dynamic architectures, implemented with a neuromorphic VLSI device for controlling simple robot behaviors: a reactive collision avoidance strategy and a spatio-temporal feature extraction to estimate relative visual motion.We demonstrate how these architectures may be realized in neuromorphic robotic agents.