Accurate 3D object detection in autonomous driving is critical yet challenging due to occlusions, varying object scales, and complex urban environments. This paper introduces the RCBEV-KAN algorithm, a pioneering method designed to enhance 3D object detection by fusing multimodal sensor data from cameras, LiDAR, and millimeter-wave radar. Our innovative Bird's Eye View (BEV)-based approach, utilizing a Transformer architecture, significantly boosts detection precision and efficiency by seamlessly integrating diverse data sources, improving spatial relationship handling, and optimizing computational processes. Experimental results show that the RCBEV-KAN model demonstrates superior performance across most detection categories, achieving higher Mean Distance AP (0.389 vs. 0.316, a 23% improvement), better ND Score (0.484 vs. 0.415, a 17% improvement), and faster Evaluation Time (71.28s, 8% faster). These results indicate that RCBEV-KAN is more accurate, reliable, and efficient, making it ideal for dynamic and challenging autonomous driving environments.