An inverse dynamics based robot control method using fuzzy identifiers

Summary form only given. In the trajectory control of robotic manipulators, the main difficulty is that the dynamics involved is coupled and nonlinear. A method for obtaining a nonlinear model is presented. To match the gravity, centripetal, Coriolis and inertial effects in the robot dynamics model, fuzzy logic systems which are represented as 3-layer feedforward neural networks are used. One of the main objectives considered is to keep the fuzzy system simple with a small number of rules and free of redundant inputs to have applicability in real time. Any deficiency in the rule base is aimed to be compensated by the fast learning capacity of the system. Firstly fuzzy modeling of the robot dynamics is considered. The 3-layer feedforward neural network representation of the class of fuzzy systems used, together with the backpropagation algorithm, are detailed. The online identification method is explained. Lastly, results are presented for the industrial arm MAMROB/ER15.