A Scalable Framework for Robust Vehicle State Estimation with a Fusion of a Low-Cost IMU, the GNSS, Radar, a Camera and Lidar

Abstract

Automated driving requires highly precise and robust vehicle state estimation for its environmental perception, motion planning and control functions. Using GPS and environmental sensors can compensate for the deficits of the estimation based on traditional vehicle dynamics sensors. However, each type of sensor has specific strengths and limitations in accuracy and robustness due to their different properties regarding the quality of detection and robustness in diverse environmental conditions. For these reasons, we present a scalable concept for vehicle state estimation using an error-state extended Kalman filter (ESEKF) to fuse classical vehicle sensors with environmental sensors. The state variables, i.e., position, velocity and orientation, are predicted by a 6-degree-of-freedom (DoF) vehicle kinematic model that uses a low-cost inertial measurement unit (IMU) on a customer vehicle. The Error of the 6-DoF rigid body motion model is estimated using observations of global position using the global navigation satellite system (GNSS) and of the environment using radar, a camera and low-cost lidar. Our concept is scalable such that it is compatible with different sensor setups on different vehicle configurations. The experimental results compare various sensor combinations with measurement data in scenarios such as dynamic driving maneuvers on a test field. The results show that our approach ensures accuracy and robustness with redundant sensor data under regular and dynamic driving conditions.