Solving Inverse Kinematics Problem for Manipulator Robots Using Artificial Neural Network with Varied Dataset Formats

Solving inverse kinematics problems presents more significant difficulties than solving forward kinematics problems, often involving greater complexity. Conventional techniques for addressing inverse kinematics problems might prove insufficient when dealing with manipulators featuring intricate joint configurations. This chapter suggests the utilization of neural computation, particularly an Artificial Neural Network (ANN), to determine the necessary joint angles for a defined Cartesian position and orientation of the end effectors for a 2-DoF manipulator. We investigate the potential of ANNs with three different datasets of joint positions: one with random step size, the other with fixed step size, and the third one is a sinusoidal signal with varying frequency to evaluate the performance of the ANN. Thereafter, input–output training datasets are created for three cases based on the direct geometrical equations. The training process of ANN is carried out by incorporating input–output datasets and compiling the MATLAB script file with tuned hyperparameters. From the numerical results, we found that the ANN technique can effectively compute the desired joint angles, a critical factor for achieving accurate manipulation control of the robotic arm. Overall, this chapter presents a novel approach to solving inverse kinematics problems using an ANN architecture with different datasets and demonstrates the potential of the same in handling intricate joint configurations.