In robot kinematics and control, it is very hard to find the solution of inverse kinematics. Conventional methods including algebraic and geometric cannot be always adequate for complex joint structures. Adaptive Neuro-Fuzzy Inference System (ANFIS) can be easier to apply and more efficient compared to these methods. The problem encountered with ANFIS usually occurs in the designing process. it includes the setting of various parameters which can be complicated and time-intensive for iterations. Facing this problem, in this paper, a Design of Experiment (DoE) methodology will be used to optimize the significant parameters of ANFIS when it is applied to inverse kinematics solution. Using Response Surface Methodology (RSM), four factors are considered as input variables. Results show that the validation error can be significantly improved using the proposed scheme. For each theta, the significant parameters were determined, and the optimal values were presented. |
