Learning-Based NMPC Adaptation for Autonomous Driving Using Parallelized Digital Twin

Abstract

In this work, we focus on the challenge of transferring an autonomous driving (AD) controller from simulation to reality (Sim2Real). We propose a data-efficient method for online and on-the-fly adaptation of parametrizable control architectures such that the target closed-loop performance is optimized while accounting for uncertainties such as model mismatches, changes in the environment, and task variations. The novelty of the approach resides in leveraging black-box optimization enabled by executable digital twins (xDTs) for data-driven parameter calibration through derivative-free methods to directly adapt the controller in real time (RT). The xDTs are augmented with domain randomization (DR) for robustness and allow for safe parameter exploration. The proposed method requires a minimal amount of interaction with the real world as it pushes the exploration toward the xDTs. We validate our approach through real-world experiments, demonstrating its effectiveness in transferring and fine-tuning a nonlinear model predictive control (NMPC) with nine parameters, in under 10 min. This eliminates the need for hours-long manual tuning and lengthy machine learning training and data collection phases. Our results show that the online adapted NMPC directly compensates for the Sim2Real gap and avoids overtuning in simulation. Importantly, a 75% improvement in tracking performance is achieved, and the Sim2Real gap over the target performance is reduced from a factor of 876 to 1.033.