Our most recent work on AI for Space features a neural network for spacecraft pose estimation, which is improved and robustified during rendezvous and proximity operations through online supervised learning using pseudo-labels generated by an adaptive unscented Kalman filter. This is all demonstrated using the most advanced hardware-in-the-loop datasets publicly available: SPEED+ and SHIRT .
Check out the short 3 min video created by Tae Ha "Jeff" Park for a full explanation.
