FNN-Based Inverse Kinematics for Efficient Trajectory Planning in Industrial Robots

Abstract

Industrial robots require precise trajectory planning for optimal performance, particularly when planning in Cartesian space, which demands solving the complex inverse kinematics (IK) problem. Traditional numerical methods for IK are accurate but slow. In contrast, Feedforward Neural Networks (FNNs) offer faster solutions but face challenges in accuracy and training time. This study examines the use of FNNs for task-specific IK with the ABB IRB 6700 robot, aiming for high accuracy and rapid training. The FNN model is trained on desired Cartesian paths and corresponding joint configurations, and its performance is evaluated using Mean Error (ME), Standard Deviation (STD), and Root Mean Squared Error (RMSE). Results show significantly reduced training times and accurate predictions of joint configurations and Cartesian poses. The FNN's predictions have lower STD values than the robot's path repeatability, demonstrating its potential to enhance real-world trajectory execution.