Spiking Neurons as Universal Building Blocks for Hybrid Systems

Abstract

Spiking neural networks in-silico can closely resemble the architecture and dynamics of neural networks in-vivo and then mimic brain functions. However, their use for applied computations remains rather limited. In this work we report two successful cases of using networks of spiking neurons for controlling mobile robots. In the first case a neural network serves as a "brain" of an animat (a crocodile toy). We show that the network can learn from the environment and reproduce basic behaviors of advancing towards an object and "biting". In the second case spiking neurons are used in a human-robot interface allowing controlling a mobile robot by hand gestures. Sensory neurons detect myographic signals from a bracelet worn on a forearm. Then, the preprocessed output is classified according to hand gestures and the corresponding command is sent to the robot. Our results show that after 3-10 trials all users manage to control the robot fluently.