LiDAR-Driven Steering Command Prediction for Autonomous Vehicles Using Machine Learning

Autonomous navigation systems rely on robust perception and control mechanisms to ensure efficient movement and decision-making. This study explores the application of machine learning (ML) models to predict steering commands using 360-degree LiDAR scans for an Arduino-based autonomous vehicle (AV). A lab-scale AV equipped with an A3 RPLIDAR sensor was used to collect about 5.4 million LiDAR scan samples in a structured indoor environment. The dataset was preprocessed through feature standardization and dimensionality reduction via Principal Component Analysis (PCA). Five ML models-Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Gaussian Naïve Bayes (GNB), and Multi-Layer Perceptron (MLP)-were trained to classify steering commands (Forward, Left, Right). During navigation, real-time LiDAR scans are processed through the trained model, and predicted steering commands are published to an Arduino-based control system for execution. Performance was evaluated using Accuracy, Precision, Recall, F1-score, and ROC-AUC, with RF achieving the highest accuracy (99.27 %) and GNB the lowest (75.52 %). The results highlight the effectiveness of LiDAR-driven ML models for realtime autonomous steering, with future work focusing on sensor fusion and deployment in dynamic environments.