A GUI-Enabled, Automated LSTM-Based Inverse Kinematics Pipeline for 6-DOF Robotic Arms Using MATLAB and CoppeliaSim

Abstract

Existing deep learning-based inverse kinematics (IK) solutions often target specific robotic arms and require significant modifications when applied to different configurations. To support early-phase design and testing of 6-degree-of-freedom (6-DOF) robotic arms, this study presents a fast and adaptable IK solution through a user-friendly interface. Unlike traditional numerical methods that are computationally intensive, sensitive to initial conditions, and may not generalize to custom designs, the proposed approach allows users to input Denavit-Hartenberg (DH) parameters and quickly generate a first-draft IK solution. This solution is built on a deep learning-based pipeline using a Long Short-Term Memory (LSTM) neural network integrated with a MATLAB-based graphical user interface (GUI) for automated dataset generation and model training. To enhance performance, this approach applies various data preprocessing techniques, including MinMaxScaler, Normalizer, RobustScaler, and StandardScaler. It also incorporates K-Fold cross-validation for performance evaluation and an early stopping mechanism to prevent overfitting. Multiple 6-DOF robotic arms are tested using MATLAB and CoppeliaSim by performing tasks, such as trajectory tracking of letters and words on planar and non-planar surfaces, to ensure a flexible solution across diverse robotic configurations and task environments.