ISA Transactions 48 (2009) 497–502 Contents lists available at ScienceDirect ISA Transactions journal homepage: www.elsevier.com/locate/isatrans New hybrid adaptive neuro-fuzzy algorithms for manipulator control with uncertainties – Comparative study Srinivasan Alavandar * , M.J. Nigam Department of Electronics & Computer Engineering, Indian Institute of Technology Roorkee, 247667, India article info Article history: Received 2 October 2008 Received in revised form 25 May 2009 Accepted 26 May 2009 Available online 11 June 2009 Keywords: Neuro Fuzzy Systems Manipulator control Uncertainties Conventional control abstract Control of an industrial robot includes nonlinearities, uncertainties and external perturbations that should be considered in the design of control laws. In this paper, some new hybrid adaptive neuro-fuzzy control algorithms (ANFIS) have been proposed for manipulator control with uncertainties. These hybrid controllers consist of adaptive neuro-fuzzy controllers and conventional controllers. The outputs of these controllers are applied to produce the final actuation signal based on current position and velocity errors. Numerical simulation using the dynamic model of six DOF puma robot arm with uncertainties shows the effectiveness of the approach in trajectory tracking problems. Performance indices of RMS error, maximum error are used for comparison. It is observed that the hybrid adaptive neuro-fuzzy controllers perform better than only conventional/adaptive controllers and in particular hybrid controller structure consisting of adaptive neuro-fuzzy controller and critically damped inverse dynamics controller. © 2009 ISA. Published by Elsevier Ltd. All rights reserved. 1. Introduction Controlling of robot manipulators has always been considered a challenging problem. Normally an N-degree of freedom (DOF) rigid robot manipulator is characterized by N nonlinear, dynamic, coupled differential equations [1–3] with uncertainty as a robot may work under unknown and changing environments and execute different tasks. The nonlinear effects have become very prominent in the presence of friction, backlash, hysteresis and elasticity of joints and links. Robot dynamics get more complicated when payload conditions and link inertia become uncertain. The most challenging problem in this field is that there are always uncertainties in the unstructured environments. Structured uncertainty is characterized by a correct dynamical model with parameter uncertainty due to imprecision of the manipulator link properties, unknown loads, inaccuracies on the torque constants of the actuators, and so on. Unstructured uncertainty is characterized by unmodeled dynamics such as nonlinear friction, disturbances, and the high-frequency part of the dynamics, neglected time- delays and so on. However, knowledge of an exact model for robot manipulators is very significant for designing efficient control algorithms. In many cases these controllers are developed under the assumption that the robot manipulators are ideal, i.e., they are not affected by nonlinear friction, elasticity or backlash phenomena [4]. * Corresponding author. E-mail addresses: seenu.phd@gmail.com (S. Alavandar), mkndnfec@iitr.ernet.in (M.J. Nigam). Some of the other controllers are developed with partial and incomplete knowledge about the robot models and most of them are concerned with the robustness of the controller, designed in the presence of uncertainties in link inertia [5]. Conventional robot control methods depend heavily upon accurate mathematical modeling, analysis, and synthesis. There are many control strategies that can be applied for control of robot manipulators. These range from conventional [6] to adaptive [7,8], to neural network [9–11], to more recent fuzzy [12–14], adaptive fuzzy control [15,16] and adaptive neuro-fuzzy [17–19] strategies. Fuzzy systems and neural networks have been successfully em- ployed over the years to map nonlinear, dynamic, multidimen- sional mathematical models. Neuro-fuzzy systems (NFSs) have been developed to make a sensible merge of linguistic information processing capability of fuzzy systems and intelligent learning ca- pability of neural networks to evolve systems which have strong modeling capability as well as relatively easy interpretability from the user point of view. To address the problem of modeling dynam- ically varying systems, Jang [17], propose an Adaptive Neuro-Fuzzy Inference System (ANFIS), in which a polynomial is used as the defuzzifier. It dis- tinguishes itself from normal fuzzy logic systems by the adaptive parameters, i.e., both the premise and consequent parameters are adjustable. Takagi and Sugeno [20] change the defuzzification pro- cedure where dynamic systems are introduced as defuzzification subsystems. The potential advantage of the method is that under certain constraints, the stability of the system can be studied. In this paper, an attempt has been made to do a comparative study of some hybrid adaptive neuro-fuzzy control schemes for the control of six DOF (Puma) robot arm with uncertainties 0019-0578/$ – see front matter © 2009 ISA. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.isatra.2009.05.003