Machine Learning for Robot Precision: Predicting End-of-Operation Errors Using Regression Techniques

摘要

This study provides an in-depth evaluation of predictive modeling techniques used to estimate end-effector error in robotic systems, comparing the performance of four machine learning models: linear regression, multilayer perceptron (MLP), decision tree regression, and support vector regression (SVR). The research focuses on addressing key challenges in precision control by analyzing how operational parameters such as joint angles, motor torque, sensor delay, and battery voltage affect the accuracy of end-effector positioning. A dataset consisting of 20 observations with joint angles ranging from 20° to 95°, motor torques ranging from 1.0 to 2.2 Nm, sensor delays ranging from 7 to 16 milliseconds, and battery voltages ranging from 22.3 to 25.1 volts was used. A rigorous testing approach was adopted using several performance metrics including R-squared (R²), mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE). Linear regression gave the best results, scoring 0.9838 on training and 0.9594 on test data, indicating a strong linear relationship between input features and end-product error. The MLP model also performed well, capturing nonlinear patterns with R² values of 0.98075 (training) and 0.94832 (test). The decision tree regression model showed a perfect fit on the training data (R² = 1.0000), while it showed overfitting, leading to a decrease in performance on the test data (R² = 0.9240). SVR provided a balanced effect with strong generalization, showing R² values of 0.9606 and 0.9501 on the training and test datasets, respectively. Correlation analysis identified strong positive correlations between the final-effector error and motor torque (0.97), joint angle 1 (0.93), and sensor delay (0.89). Conversely, battery voltage had a strong negative correlation (-0.95), indicating that higher voltage improves position accuracy. These insights underscore the importance of efficient power usage and timely sensor feedback in achieving high-precision robot control