Integrating Advanced Machine Learning, Sensor Fusion, and Control Systems to Enhance Autonomous Vehicle Safety and Performance

Self-driving vehicles are right now one of the most progressive elements in the contemporary automotive industry, they may bring significant improvements for traffic safety, traffic intensity, and emission levels. However, reaching the necessary level of operational reliability for masse implementation implies surmounting issues with real-time occupancy decision making, precise localization, and accurate object identification. This paper presents a systematic approach to these problems using Machine Learning algorithms, Sensor Fusion, and Control Systems. Namely, the framework assesses CNNs and RNNs in core AV functions based on detection, classification, and trajectory prediction capabilities. In CNN evaluation arises with a classification of $94 \%$ and $F 1$ score was $\mathbf{9 4 \%}$ In the similar classification, RNN has a minimal error of $0 \%$ with overall accuracy of $\mathbf{9 1 \%}$ and precision of $\mathbf{9 0 \%}$. For increasing the accomplishment of localization and perception, different sensor fusion techniques were discussed and based on the conclusion that the EKF offered improved performance; the proposed approach achieved accurate localization up to 0.20 meters and an object detection of percent 94. Based on these results, an improved cost function was formulated, and an MPC solution was thus applied. This approach provides a tracking error which was reduced to 0.10 meters, and it has excellent performance compared to the basic PID and LQR control in similar circumstances. In doing so, this work lays down a strong foundation for further research on AV technology and the relevance of human expertise on future $A V$ systems safety and efficiency.