On-board recursive state estimation for dead-reckoning in an autonomous truck

Abstract: Most of fully-autonomous vehicles are equipped with GPS devices in order to keep track of their exact location while driving towards any target destination. However, it is widely known that GPS systems are likely to fail under certain conditions, e.g., in long tunnels or during very bad weather conditions. In this master thesis work we present an Extended Kalman filter (EKF) framework for dead-reckoning in au-tonomous trucks equipped with a CPU, a gyroscope and four simulated sensors: a GPS, a magnetometer and two rotary encoders for velocity. The EKF will fuse the sensors measurements with a prediction that uses the kinematic model of a non-holonomic vehicle. In order to improve the estimation of the yaw angular position when a GPS outage is reported a new calibration method based on the rotation matrix is applied. This method is proven to e˙ectively decrease the error while driving in GPS denied environments.The tests are performed in a real-time embedded system, NI myRIO, that runs on-board of a 1:14 scaled Scania truck. The performance results confirm the correctness of our framework under short-term GPS outages, during many driving loops. Additionally, during long-term outages the estimation works pretty well for one loop and it has a good performance for multiple loops due to the unavoidable sensor drifting.