At Intel, we use our high-volume semiconductor manufacturing expertise and 300 mm processes to fabricate large two-dimensional quantum dot arrays for exchange-only spin qubits. As arrays increase in size, manual tune-up stops being practical. Measurement automation is a requirement for operating these devices at scale. In this talk, I will discuss the challenges of automating the bring-up and calibration of large 2D arrays. These include setting the desired electron occupation across many coupled dots, compensating for crosstalk, and organizing the many interdependent calibration routines into a scalable framework. I will present our recent progress toward tuning up 2D devices from initial cooldown to qubit operation, and how we benchmark progress on our crossbar device, an 11x11 array of qubit devices. A central part of this workflow is machine-learning-based analysis of measurement data, such as charge stability diagrams and qubit characterization scans. Models trained on simulated or historical data often degrade when applied to new devices, cooldowns, or measurement conditions: a problem called domain shift. I will show how training neural networks to ignore superficial differences between datasets (a domain adversarial approach) helps them learn features that stay consistent across these varying conditions. This makes automated decisions more reliable on new, previously unseen devices and reduces the need for costly labeling. Together, these techniques form the basis for scalable, automated control of large 2D spin qubit arrays.