|A difficult goal in robotics is solving the inverse kinematics for N degree-of-freedom (DOF) robotic arms. Solving for the joint angles for a robot to follow a specified trajectory is a complex task as this requires solving for the solution of an end-effector with respect to its links and base coordinates. Depending on the number of DOFs, finding a closed form solution becomes impossible to find and computationally expensive. An alternative to this consists of function approximations in the form of neural networks (NN). Neural networks have the utility of mapping functions to inherently nonlinear processes such as solving for the trajectories of complicated manipulators through an iterative process of updating weight coefficients based on cost functions, error criteria, and optimization algorithms. In this paper, we propose a method that combines several different NN models along with a method for specifying different batches from the training set based on a set of error criteria with the ultimate goal of improving accuracy.|
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.