MT3D-Seg: Unified Multi-Task Learning Framework for 3D Object Detection and Drivable Area Segmentation in Smart Mobility

Robust perception is essential for ensuring the safe operation of autonomous systems in smart mobility. Multitask learning (M-TL) enhances perception by simultaneously addressing multiple tasks, improving efficiency, and optimizing performance. This paper introduces MT3D-Seg, a novel MTL framework that integrates 3D object detection and drivable area segmentation, with a primary focus on 3D object detection. Trained on the KITTI dataset, MT3D-Seg leverages shared feature representations to handle complex road scenarios effectively. Experimental results demonstrate high accuracy across both tasks while reducing computational overhead, making it a promising solution for real-time environmental perception in autonomous driving.