Disturbance-distribution learned Tube MPC for current-command gimbal pointing control
The disturbance distribution is learned and used to tighten Tube-MPC constraints online. On a two-axis current-command gimbal (Intel N100 + CAN, 100 Hz), the 2-D pointing RMSE is 3.94 mrad, about 47% lower than nominal MPC, with zero current-limit activations and a worst-case solve time of 4.5 ms.