I had assumed the purpose of the data was more in generalising across variegated input sources to better allow the drones to fly on their own in urban settings, aka, adapt more readily to randomness? Better datasets for multimodal training etc.

I am not joking though, I really would consider any data generated on public assets to be considered "releasable" to the public. How many people should get killed by self-driving cars because the company making the cars didn't have enough data to train proper models?