Inverse Kinematics for A 3-R Robot Using Artificial Neural Network and Modified Particle Swarm Optimization

Abstract

As the calculation of the exact position and orientation of the end effector of robot manipulator is mandatory to obtain inverse kinematics solution, an artificial neural network is used to obtain inverse kinematics as it reduces the computational time as well as complexity associated. The backpropagation algorithm that is generally used for updating weights and biases requires the sensitivity function of the system which is sometimes difficult to obtain. Here, training of neural network means optimizing the parameters of neural network, i.e. weights and biases, for which particle swarm optimization (PSO) is used. PSO is suitable for learning neural network as it does not require the derivative of an objective function. In this work, the parameters of neural networks, i.e. weights and biases, are optimized using three different optimization algorithms, i.e. PSO, segmented particle swarm optimization (SPSO) and modified segmented particle swarm optimization (MSPSO), and their results are then compared. Further, a comparison has been made on the basis of two parameters, i.e. fitness function and regression, and MSPSO is found to perform better than the other two optimization algorithms. Fitness function value obtained using MSPSO is 6.18e−09, using SPSO is 1.1e−05 and using PSO is 2.35e−03.