Abstract
Sustainability is the success factor of the industry 5.0 era, where industries are focused towards customer-centric development. The exponential growth of smart cities paves way for opportunities for the development of various automated customer centric developments. Automation is the backbone of sustainable smart city development. The proposed work is one such sustainable solution which provides for the usage of Autonomous Electric Vehicles (AEV) for driver-free vehicle operation. This proposed research presents a groundbreaking approach to AEV that addresses the unique challenges of unstructured roadways in developing countries and smart cities. With the integration of the multiple deep learning models in a cascaded architecture, this work creates a comprehensive system capable of handling the diverse and challenging road conditions found in countries like India. The core innovation lies in the unified framework that simultaneously processes lane boundaries and critical objects at 6 frames per second on resource-constrained hardware, with intelligent prioritization of safety features. Performance metrics are exceptional with measures of 97.26% accuracy for lane detection using DeepLabv3+, 0.92 mAP for object detection with YOLOv5, and 0.83 mAP for pothole detection using YOLOv7. The successful implementation on a custom-built electric vehicle platform demonstrates the commercial viability of this approach, potentially bridging the adoption gap for autonomous technology in developing economies worldwide.