Control of a One-Legged Hopping Robot using a Hybrid Neuro-PD Controller

Abstract

Primary simulation results of the control of a pneumatically actuated hopping robot along with the mathematical model are presented in. This paper presents the next phase of the research: design of a robust controller for an experimental hopper. Dynamic stability of the hopping robot is investigated using an artificial neural network (ANN)-based proportional-derivative (PD) controller. The hopper's model (i.e. the transfer function of the plant) is identified with the help of an ANN, and then the PD controller is integrated with the trained ANN, so that the plant's output follows a pre-specified reference trajectory. It is evident through computer simulations and experimental results that the proposed controller effectively meets the system's performance requirements, i.e. achieving a user-defined constant jumping height after a number of hops. It is noteworthy that a near zero steady state error and a shorter settling time in the presence of unmodeled system dynamics can be achieved by incorporating an inverse dynamics paradigm into the proposed PD controller in conjunction with an ANN