There are many different applications. I worked on close proximity manoeuvres where we were using deep learning for pose estimation, you can do something similar for Lunar/Mars landing. You can also pre-process your data onboard to save on the data budget. And you can also do analytics in real time.

Interesting, thanks.

And for orbital mechanics, yes we can do it. But when something goes wrong (thruster malfunction, error in pointing), it can be complex to recover in real time with traditional algorithms. Given the difficulties in sat comms, this meant a lot of effort was put on making sure things go right 1/2