Failure Detection Under Sensing Uncertainty in Vehicular Systems

The advancement of autonomous vehicles depends on the integration of sophisticated sensors that perceive their environment in real-time. However, these sensors are prone to uncertainties that, if not properly managed, could undermine the vehicle’s decision-making process and elevate road safety risks. In response to this challenge, we present an embeddable solution designed to filter uncertain data using a machine learning model. Our solution not only resists to inconsistencies in sensor data but also quantifies the degree of anomaly in sensor outputs. This capability is essential for enabling effective predictive maintenance by identifying potentially failing systems before they compromise the overall performance. Through simulations, we demonstrate that the anomaly system detector enhances the vehicle’s resilience against discordant data.