An Optimized Uncertainty Aware Sensor Fusion approach for Autonomous Vehicles using Bayesian Neural Networks

Autonomous vehicle is a one of the most emerging technologies that have started being used in developed countries for public transportation, such as robotaxis, autonomous buses. During uncertain situations, such as heavy rain, mist and bad weather the reliability of the sensor values may compromise due to decreased visibility, increased sensor noise, and inconsistencies in sensor reading. In this paper, uncertainty aware sensor fusion approach is proposed based on Bayesian Neural Networks (BNN) to address this problem. Uncertainty estimation and redundancy graph construction provides the stability to analyze the sensor data readings. It also enhances the accuracy and robustness in state estimation. An adaptive confidence weight helps the system to prioritize the sensors and the graph structure ensures the propagation of probabilistic data’s across the nodes. The simulation result demonstrates the improved robustness and reliability in the multi sensor fusion process and better accuracy in the state prediction and decision making process.