<p>Self-supervised monocular depth estimation offers a practical solution for autonomous vehicles by eliminating dependency on ground truth data in training. However, existing methods often prioritize accuracy at the cost of computational efficiency, limiting their deployment on resource-constrained edge devices. To address this gap, we propose a novel hybrid architecture combining a simplified transformer with a convolutional neural network (CNN). The development and training of such models necessitate high-performance computing (HPC) resources for large-scale data processing and rapid experimentation. Our approach integrates a simplified transformer with a CNN to build a hybrid model that reduces computational complexity while retaining robust spatial reasoning capabilities. This approach enables a compact model design that significantly reduces computational cost compared to conventional approaches. For autonomous driving application, we validate our framework on the KITTI dataset, demonstrating competitive depth estimation accuracy against state-of-the-art methods. By achieving strong performance with reduced computational demands, our work provides a viable pathway for deploying efficient depth estimation models in real-world autonomous vehicle systems.</p>