Hybrid Algorithm for Inverse Kinematics Using Deep Learning and Coordinate Transformation

This paper presents a novel algorithm to solve the inverse kinematics (IK) problem for a complex robotic manipulator with a high degree of freedom (DOF). Our proposed algorithm divides the manipulator's joints into two sets and employs deep learning (neural network (NN)), a coordinate transformation, and vector analysis to solve the IK problem as two sub-problems. Our approach has low computation complexity, especially after training the NN, and it does not require imposing any simplifications of the complex robot structure to simplify the corresponding IK problem. We apply our approach to solve the IK problem for the Baxter robotic manipulator and demonstrate results enabling the Baxter's end-effector to reach desired positions and orientation with a small amount of error.