Search Results

This paper presents the design and experimental validation of a localization method for autonomous driving. The investigated method proposes and compares the application of the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) to the sensor fusion of onboard data streaming from a Global Positioning System (GPS) sensor and an Inertial Navigation System (INS). In the paper, the design of the hardware layout and the proposed software architecture is presented. The method is experimentally validated in real time by using a properly instrumented all-wheel-drive electric racing vehicle and a compact Sport Utility Vehicle (SUV). The proposed algorithm is deployed on a high-performance computing platform with an embedded Graphical Processing Unit that is mounted on board the considered vehicles.

Electric vehicles comprising multiple motors allow the individual wheel torque allocation, i.e. torque-vectoring. Powertrain configurations with multiple motors provide additional degree of freedom to improve system level efficiencies while ensuring handling performances and active safety. However, most of the works available on this topic do not simultaneously optimize both vehicle dynamic performance and energy efficiency while considering the real-time implementability of the controller. In this work, a new and systematic approach in designing, modeling, and simulating the main layers of a torque-vectoring control framework is introduced. The high level control combines the actions of an adaptive Linear Quadratic Regulator (A-LQR) and of a feedforward controller, to shape the steady-state and transient vehicle response by generating the reference yaw moment. A novel energy efficient torque allocation method is proposed as a low level controller.

Modern vehicles have several active systems on board such as the Electronic Stability Control. Many of these systems require knowledge of vehicle states such as sideslip angle and yaw rate for feedback control. Sideslip angle cannot be measured with the standard sensors present in a vehicle, but it can be measured by very expensive and large optical sensors. As a result, state observers have been used to estimate sideslip angle of vehicles. The current state of the art does not present an algorithm which can robustly estimate the sideslip angle for vehicles with all-wheel drive. A deep learning network based sideslip angle observer is presented in this article for robust estimation of vehicle sideslip angle. The observer takes in the inputs from all the on board sensors present in a vehicle and it gives out an estimate of the sideslip angle. The observer is tested extensively using data which are obtained from proving grounds in high tire-road friction coefficient conditions.