GA-based control to swing up an acrobot with limited torque

Abstract

To swing up an acrobot, a two-link under-actuated robotic system, a fuzzy neural network (FNN) controller with limited torque based on genetic algorithm (GA) is presented and tested. The acrobot is composed of two links rotating in a same plane around two joints, only the second of which is actuated. The upswing process is a complex problem to rotate the two links from the low to the top balance position by single torque on the actuated joint. In a real application system, motor output is often limited. Therefore, an FNN controller is proposed to generate limited torque on the actuated joint according to different states of the two links. The antecedents and consequences of the FNN controller are determined with an improved GA. Simulation results on the acrobot system verify the feasibility and generalization of the proposed GA-based FNN control scheme.