Autonomous Driving: Object Detection, Path Planning, and Decision Making using Deep Learning Model

In this paper there is an architecture implementable for fault-tolerant pipelines for autonomous driving systems along with path planning, object detection and decision making. The proposed solution is an intermediate-level solution combine several deep learning algorithms in the hierarchical structure, thus improving the reliability and the overall performance of the system. The pipeline begins with generating input data, such as images and videos from dashboard cameras and network streams, and encompasses multiple processing steps up to risk assessment and risk mitigation techniques. For fault tolerance, the architecture relies on fallback techniques since if one algorithm fails, another algorithm prescribed through configuration will be successfully taken over. Road segmentation is achieved using FCNs, but U-Net can also be enabled if necessary. It dynamically deploys YOLO and Single Shot MultiBox Detectors (SSD) for 2D object detection, which allows for algorithm switching corresponding to performance conditions. The object tracking module thus combines Multiple Object Tracking (MOT) algorithms like Deep Sort, Byte Track, and Bot Sort to support accurate and consistent object tracking in dynamic surroundings. To attain thorough situational awareness, the system also uses 3D object detection based on LIDAR. The path planning module ensures safe and effective route navigation by combining well-known algorithms like A*, Dijkstra's Algorithm, and its variations. The multi-layered architecture of the suggested pipeline improves autonomous driving systems' resilience by offering a dependable framework that can preserve operational integrity in a range of circumstances.