Recurrent Neural Networks for Real-Time Position Estimation of Robotic Arm End-Effector Coordinates

Robotic arm manipulators find application in fields like manufacturing, construction, and automation due to their inherent ability to perform accurate and repetitive operations. These manipulators consist of multiple joints and links, enabling complex, multi-directional motions. However, accurately determining end-effector positions poses significant challenges due to the complexity of kinematic calculations. Due to the difficulty in estimating the formula to get to this solution, we have proposed the addition of a Deep Learning Model based Recurrent Neural Networks (RNN) to predict the end-effectors X, Y, and Z positions according to the input of the 6 joints of the robotic arm. A robust dataset is generated using the Denavit-Hartenberg (D-H) parameterization for randomly sampled joint configurations. The RNN model is trained and validated over 50 epochs using a dense layer with 100 units and subsequently evaluated using three key metrics: Mean Absolute Error (MAE), Mean Squared Error (MSE), and R-Squared Score. Enhancements to the analysis of the models performance will include confusion matrices and Receiver Operating Characteristic (ROC) curves to assess the accuracy of each category. Finally, the developed model predicts the end-effector for any user input with high accuracy, providing an efficient and reliable solution for real-time kinematic computations in robotic applications.