DrivNet: Adaptive driving behavior prediction for improving the safety of multimodal transportation in autonomous vehicle systems

The safety and efficiency of autonomous vehicles depend on their ability to react swiftly and appropriately to dynamic driving environments. This paper presents DrivNet (Adaptive Driving Network), an adaptive algorithm, DrivNet, aimed at enhancing the reaction time and decision-making capabilities of autonomous vehicles under varying traffic conditions and times of day. DrivNet integrates vehicle parameters and driving behavior with physiological features to improve forward movement within situational awareness environments. The algorithm is validated using a real-time driving dataset collected from expert drivers, capturing both vehicle and driving behavior data under heavy and normal traffic conditions, as well as day and night scenarios. Distinct driving behavior patterns are generated for three key situational awareness conditions: accelerating on clear roads, navigating through critical situations, and maintaining fuel efficiency. These patterns serve as the basis for adapting vehicle control decisions. To validate the effectiveness of these driving behavior patterns, a Recurrent Neural Network (RNN) architecture is employed, enabling the detection and classification of psychological features such as mental workload, stress, and fatigue. The proposed DrivNet algorithm offers valuable insights into distinguishing safe from unsafe driving modes, thereby supporting an intelligent control mechanism that enhances the overall safety of autonomous transportation systems. The results demonstrate the potential of DrivNet to improve autonomous vehicle performance, contributing to the future of safe and efficient self-driving technologies.