2024 VI-GRADE ZERO PROTOTYPES SUMMIT

In our presentation, we unveil a transformative strategy: leveraging VI-WorldSim for synthetic semantic segmentation to bolster autonomous perception and adopting VI-CarRealTime to ease the end-of-line-tuning of vehicle controllers and observers with Twin-in-the-Loop architectures. The fusion of VI-WorldSim and synthetic data represents a paradigm shift in autonomous driving technology, offering scalable and adaptable solutions. This innovative approach not only improves scene understanding but also streamlines deep learning model training, expediting the development process of autonomous driving systems. Additionally, the incorporation of synthetic data enhances the safety of the perception stack for real-world deployment, mitigating risks associated with unreliable or inadequately trained models. These advancements hold significant promise for practical applications, notably in the upcoming 2024 edition of the 1000 Miglia race, where our team plans to employ this approach for navigation throughout Italy. Our presentation highlights the pivotal role of the mentioned VI-grade solutions in driving progress in the field, underscoring its contribution to the development of perception algorithms tailored for real-world scenarios. In addition, another significant advancement in the automotive field lies in the integration of Twin-in-the-Loop (TIL) architectures. These systems directly embed Digital Twins into the loops, eliminating the need to calibrate control-oriented models with time-consuming experimentation. TIL controllers require minimal real-life experiments, relying mostly on simulated data, while TIL observers enhance precision by unifying vehicle state estimation, marking a new era of automotive innovation.TIL controllers ease the end-of-line tuning process and reduce the deployment time of even the most sophisticated and complicated controllers. Instead of conducting expensive experiments on the real vehicle, the controller is calibrated solely in simulation. When deployed, it closes two simultaneous loops on both the physical vehicle and its digital twin, adding with a simple compensator to account for modeling mismatches. The counterpart to the controller is the Twin-in-the-Loop observer, capable of estimating all vehicle states using a single comprehensive vehicle simulator. Instead of employing numerous simplified vehicle state observers for each dynamic of interest, the aim of the TIL observer is to unify all vehicle estimation problems using a single digital twin. The TIL observer outperforms conventional simplified dynamics observers by incorporating typically unmodeled dynamics and coupling effects. This convergence of physical and digital realms not only streamlines development but also enhances precision, heralding a new era of automotive innovation.