Improving Suspension Control in Autonomous Vehicles Using Camera-Based Pothole Geometry Detection and a Quarter-Car MPC Approach

Active suspension systems can reduce the impact of road irregularities on vehicle comfort, but traditional controllers only react after the disturbance occurs and do not exploit preview information from on-board sensors. In this paper, we propose a camera-based approach that uses a YOLOv8 model to detect potholes ahead of the vehicle. The full pothole geometry, namely width, longitudinal distance, and depth, is estimated from monocular images using a pinhole camera model combined with a calibrated empirical regression. This geometric information is used to parameterise a half-sine (versine) road disturbance profile, which serves as a preview input for a Model Predictive Control (MPC) strategy applied to a quarter-car suspension model. The preview MPC adjusts the suspension force proactively, while an IoT MEMS accelerometer with an Extended Kalman Filter (EKF) provides online state correction. The complete pipeline is implemented and evaluated in MATLAB simulation. Results show a 28.9% reduction in peak body acceleration and an 8.3% reduction in RMS body acceleration compared to a passive suspension, and a 9.3% improvement in peak body acceleration over a reactive MPC without preview.

Hali tarjima qilinmagan