The Twin-in-the-Loop approach for vehicle dynamics estimation and control: theory and applications
In vehicles, the ever increasing computing power in on-board Electronic Control units enables the use of demanding software components. Specifically, vehicle Digital Twins are full-fledged dynamic simulators, used for off-line design and testing: what would be the impact of embedding such sophisticated models in on-line estimation and control algorithms? The developed framework is denoted as Twin-in-the-Loop (TiL).
On the control side, in typical design frameworks a certain controller (possibly a very sofisticated one) is finely tuned on increasingly complex vehicle models, up to a final implementation on the real vehicle, requiring a long and constly End-of-Line calibration: we change this framework by directly employing the controlled simulator on the real vehicle as an open-loop feedforward compensator, while a second simpler loop is enforced to guarantee stability of the overall system.
On the estimation side, recent research showed as TiL observers (employing a vehicle simulator as a plant replica) are able to outperform benchmark ones: a critical aspect when considering TiL estimators also lies in the calibration of the algorithm itself. Such algorithm is in fact characterized by many hyperparameters, while also being a black-box, thus preventing the use of classical observer tuning approaches (e.g. the Kalman Filter theory).
Back to Alumni