Designing a Layered Learning System and Layer by Layer Implementation of a Controller on Quad-rotor Based on Growing Particle Swarm Optimization

In the paper an auto-pilot with layered control architecture has been designed for an autonomous quad-rotor and an optimized proportional-derivative controller has simulated on the quad-rotor. In the auto-pilot with layered control architecture, quad-rotor will be controlled with a multi agent system which has at least 4 agents to control the system instead of just having a single controller. Each agent has placed in an independent layer with its own CPU, memory and processing equipment that perform individually. Nonlinear controllers such as back stepping and sliding mode are presented. In the layered learning process, a new optimization method for neural networks is presented in the paper which is named Growing Particle Swarm Optimization. In one of the layers some classic controllers such as back stepping and sliding mode is used to train the neural controller in the other layer. Optimization results show that the neural network controller can successfully learn the controlling task using Growing Particle Swarm Optimization.