Safety-critical data-enabled predictive control for wheeled mobile robot

This paper addresses the real-world, safety-critical planning problem for a wheeled mobile robot with an unknown model, focusing on obstacle avoidance in complex environments. Data-enabled predictive control (DeePC) is a renowned optimal control framework for unknown linear time-invariant systems, offering the ability to predict future system trajectories using only past input/output measurements, without requiring explicit system identification. To enable its application to nonlinear systems, a local linear approximation technique with online data updates is adopted. To further improve robustness, a novel adaptive Lasso-based regularization term is incorporated into the cost function. Safety is a critical challenge for data-enabled control of nonlinear systems, which is addressed by incorporating discrete-time control barrier functions into a quadratic programming-based safety filter operating alongside the DeePC. We provide a formal proof of the practical stability of the closed-loop system, ensuring bounded trajectory tracking. Real-time simulations in the ROS-Gazebo environment show that adaptive regularization reduces the trajectory tracking error by about 20.2%.