Science and engineering is being revolutionized by computational advancements in machine learning and AI algorithms.   Leveraging sensor measurements, often in combination with scientific computation proxies, such algorithms aim to learn effective models for a diversity of downstream tasks, including reconstruction, forecasting, design and control in challenging environments that include noisy measurements and or parametric variability.  A grand challenge in the deployment of such algorithms is the fact that many physical systems are not amenable to full state measurements, but rather only discrete point sensor measurements at prescribed and limited locations.  In fact, the only knowledge of the full state space is typically acquired by simulations of the approximating governing equations through a diversity of low- or high-fidelity models.  This work aims to integrate scientific computing, data assimilation and physics informed AI in order to demonstrate tractable pathways forward for closing the SIM2REAL (simulation to reality) gap in science and engineering, thus enabling AI models that are accurate in reality, produce models that respect the fundamental laws nature, and are trustworthy in practice.