The development of autonomous vehicle (AV) technology is significantly dependent on sensor systems to obtain accurate and reliable perception of the environment. The current paper examines the integration of multi-model data fusion, here visual and audio data, into AV perception system performance improvement. The research examines the difficulties of single sensor modelities, especially in edge scenarios like occlusions, bad weather, or poor visibility, and suggests a multi-model fusion strategy to overcome such challenges. Utilizing Automated Machine Learning (AutoML) methods, the system tunes the fusion model to enhance accuracy, eliminate false negatives, and enhance precision for infrequent events. Experimental findings show that the fused model performs better compared to visiononly and audio-only systems, showing a strong decrease in false negatives and a 12% boost in precision for identifying rare objects, including emergency vehicle sirens. The fusion system also achieves real-time processing needs with a total latency of 32 ms. Robustness testing also reveals that the fusion model works consistently even in noisy environments. This research highlights the advantages of multi-sensor fusion and AutoML for autonomous vehicle systems and presents a path toward more resilient and flexible AV perception capabilities.