Deep Learning in Autonomous Vehicles for Dynamic Decision-Making: Enhancing Safety and Navigation

Self-driving cars (SDCS) are beginning to selfdisrupt the transport industry through the guaranteed improvement of safety, speed, and comfort. However, the volatile and complex road conditions present great difficulties to the decision-making process of such systems. This research work outlines a new deep learning framework for real-time decision making for AVs that encompass perception, behavior prediction, path planning and control of motion. The described framework also utilizes modern methodologies, specifically, YOLOv5 to detect objects or events of interest, RNNs with attention models to predict the dynamic behavior, PPO with A* algorithms for path planning, and motion control. These are proved by comprehensive evaluation and analysis of applicability and scalability of the system in a variety of driving contexts such as urban traffic, rural roads, highways, and unfavorable weather conditions. The perception module was able to accomplish a detection accuracy of 96.4% further minimizing latency and the behavioral prediction module was 91.8% accurate in terms of trajectory prediction. Implementation of path planning for autonomous vehicles concluded that fuel efficiency and collision avoidance rates were on higher level in order to provide comfortable journey and safety to passengers. This study underlines the future of deep learning in AV systems with interactive, responsive, and safe operational decision making in real environments. Through identifying major milestones in navigation and safety, this work advances the subsequent generation of AV solutions to formulate their societal utility.