Adaptive Data-Driven Modeling and Tracking Control of Autonomous Vehicles using Deep Bilinear Koopman Method

This study addresses the problem of data-driven modeling and tracking control for autonomous vehicles with unknown parameters. We use Koopman theory and deep neural networks to approximate vehicle dynamics, learning a finitedimensional bilinear Koopman operator from offline data. An adaptive method is proposed to continuously adjust the Koopman model based on online deviations. The bilinear model is then used to design a model predictive controller for trajectory tracking. Numerical simulations validate the method's effectiveness in adapting to uncertainties and tracking predefined trajectories.