Dynamical control by recurrent neural networks through genetic algorithms

Abstract

SUMMARYIn this study we composed a recurrent neural network learning controller and applied it to the swinging up and stabilization problem of the inverted pendulum. A recurrent neural network was trained by a genetic algorithm which had an internal copy operator or inter-individual copy operator. An appropriate controller was acquired in a recurrent neural network by training with a simple evaluation function. The recurrent neural network acquired two completely di!erent rules for swinging up and stabilization of a pendulum. It outputted these two rules continuously so that swinging up and stabilization of a pendulum was realized. Internal copy and inter-individual copy accelerated learning e!ectively by copying a part of a chromosome.

Hardware spiking neural network prototyping and application

“…Inspired by biology, researchers aim to implement reconfigurable and highly interconnected arrays of neural network elements in hardware to produce powerful signal processing units [5][6][7][8][9][10][11][23][24][25]. Execution architectures for SNN neural computing platforms can be broadly categorised as software-based (multi-processor) [10,12], FPGA [5,6,8,26] or analogue/mixed-signal [7,9,11,27,23].…”

“…Pasero et al [36] and Uribe et al [37] have succeessfully implemented an FPGA-based inverted pendulum hardware SNN controller using back-propagation and reinforcement-learning ANNs, respectively. Similarly, several researchers have demonstrated the feasibility of applying evolutionary techniques to locate an inverted pendulum controller solution [24,25].…”