Typical testing process of ADAS algorithms (perception, control, AI…) follows the X-in-the-Loop (XiL) approach where real driving data is limited used in early stage development. This talk discusses a Digital Twin framework for autonomous driving testing and validation (DriveTwin) that enables combining simulation and real vehicle testing data. Via deploying executable Digital Twin (xDT) models of vehicle dynamics, sensors and traffic in parallel with vehicle driving, the DriveTwin framework helps to accelerate algorithm calibration & validation, detect issues early on by exposing algorithms to real traffic data, capture corner cases and populate relevant virtual scenarios. The testing process will be more time and cost efficient; moreover, it allows to exploit logged driving data. Some ADAS use cases and demonstrations will be shown: scenario detection and generation, imitation learning, sim2real control and ADAS comfort testing.