Predicting Steering Angles for Autonomous Vehicles Using Machine Learning

Autonomous vehicle navigation relies heavily on accurately predicting steering angles based on a vehicle's waypoints and heading angles. Integrating waypoints, relative distances, and vehicle orientation is crucial in enabling precise control during navigation. This paper proposes a machine learning-based approach using multiple regression models to predict steering angles for autonomous vehicles, utilizing data generated from the Carla simulator. The dataset includes x and y coordinates of the vehicle's current position, target waypoints, and heading angles collected from various towns in the simulation environment. The relevant features were extracted using decision trees, and feature correlations were visualized through heatmaps and pairwise plots. We evaluated three regression models—Random Forest Regression, K-Nearest Neighbors, and Elastic Net Regression. Among these, Random Forest Regression performed best with an R -squared value of 0.762763, an MSE of 0.02840, and the lowest error metrics. These results suggest that the RFR model is the most effective in capturing nonlinear relationships between waypoints and steering angles, providing a robust solution for steering angle prediction in dynamic driving environments.