Sim2real for Autonomous Vehicle Control using Executable Digital Twin

In this work, we propose a sim2real method to transfer and adapt nonlinear model predictive controller (NMPC) from simulation the real target system based on executable digital twin (xDT). The xDT is high fidelity vehicle dynamics simulator, online in control parameter randomization learning process. parameters are adapted gradually improve performance deal with changing real-world environment. particular, metric not required be differentiable nor analytical respect necessary linearized. Eventually, proposed framework leverages altogether data-driven estimations, optimization efficiently developed environment platform. Our experiment demonstrates that achieved without tedious time labor consuming tuning.