Implementing a Neural Network System to Solve the Inverse Kinematics of a Biologically Inspired Robotic Cat Leg

Abstract

A neural network capable of solving the inverse kinematics of a four degree of freedom biologically inspired robotic cat leg (qualified as a serial linkage system) within its effective 3-D workspace is presented in this paper. The workspace consists of layers of similar but highly nonlinear cells whose vertices are associated with known kinematic variables provided by the robotic leg. The proposed neural network uses geometric properties coupled with the desired end effecter location as the neural network inputs to locate the cell for which encapsulates the associated location. Another neuron layer utilizing activation functions trained with the Perceptron Fixed learning rule is applied to interpolate within the identified cell. The similarity associated between all of the cells allows the trained neural network to effectively be applied in solving the inverse kinematics of the entire workspace.