Sampled-Data Neural Network Observer for Motion State Estimation of Full Driving Automation Vehicle
Résumé
The lateral state information of vehicles is pivotal for their lateral control and safety monitoring of motion states. In light of the high cost associated with a vehicle's lateral speed sensor, this study focuses on estimating the lateral velocity of autonomous vehicles using the vehicle's longitudinal speed and yaw rate. The main difficulty is that the state estimation is affected by sampled and delayed sensor data measurements as well as unknown modeling uncertainties. To overcome these challenges, this article proposes a novel sampled data neural network observer. The proposed observer consists of two components: a continuous state observer and a compensating injector. The compensating injector is devised to offset the information loss during sampling by digital sensors. Additionally, a radial basis function neural network is employed to approximate the unknown dynamics and modeling uncertainties in the vehicle systems. The weights of the neural network are updated using a newly designed weight update law. Furthermore, to tackle the challenge of both sampling and delay measurement, an additional integrator is incorporated in the designed compensating injector. This additional integrator aims to mitigate the impact of sensor data delays on the estimation process. Finally, the stability of the proposed method is demonstrated using the Lyapunov technique. The effectiveness of the proposed sampled data neural network observer is verified through simulation and experiment performed on a full driving automation vehicle experimental platform.