A solution to the end-effector position optimisation problem in robotics using neural networks

Abstract

The paper presents an original application of the Hopfield-type neural network to a robotics optimisation problem. The robot considered features an arm composed of three revolute joints, the last of which is the end-effector. The robot is planar as the movement of the end-effector is limited to one plane. The mechanical characteristics of the actuators in the joints, the accuracy of the angle position sensors, and dimensional errors in the mechanical elements which make up the end-effector all contribute towards an end-effector positioning error along an assigned trajectory. The computational complexity of the algorithmic solution to the minimisation of this error is at times incompatible with certain particularly critical industrial applications. To reduce the calculation time, the author presents a neural approach based on a Hopfield-type model. A detailed definition of the neural approach is given, its capacity for solving the problem is demonstrated, and the computational complexity is analysed. This analysis shows the drastic computational reduction provided by the neural approach as compared with an algorithmic solution to the problem of end-effector position optimisation.