Neuro-fuzzy control applied to multiple cooperating robots

Purpose - The paper aims to advance methodologies to optimize fuzzy logic controller parameters via neural network and use the neuro-fuzzy scheme to control two cooperating robots. Design/methodology/approach - The paper presents a special neural network architecture that can be converted to fuzzy logic controller. Concepts of model predictive control (MPC) have been used to generate optimal signal to be used to train the neural network via backpropagation. Subsequently, a trained neural network is used to obtain fuzzy logic controller parameters. Findings - The proposed neuro-fuzzy scheme is able to precisely learn the control relation between input-output training data generated by the learning algorithm. From the experiments performed on the industrial grade robots at Robotics and Manufacturing Automation (RAMA) Laboratory, it was found that the neuro-fuzzy controller was able to learn fuzzy logic rules and parameters accurately. Research limitations/implications - The backpropagation method, used in this research, is extremely dependent on initial choice of parameters, and offers no mechanism to restrict the parameters within specified range during training. Use of alternative learning mechanisms, such as reinforcement learning, needs to be investigated. Practical implications - The neuro-fuzzy scheme presented can be used to develop controller for plants for which it is difficult to obtain analytical model or sufficient information about input-output heuristic relation is not available. Originality/value - The paper presents the neural network architecture and introduces a learning mechanism to train this architecture online. © Emerald Group Publishing Limited.

Duke Scholars

Author Devendra P. Garg Thomas Lord Department of Mechanical Engineering and Materia ...