I am currently working on a robot MPC application. The SQP_RTI solver can successfully solve at the very first round. Then I use the state at the end of the first step as the initial state of next solving, it reports error status 3 from HPIPM after several iterations of qp. However, if I reload the solver at each round, the SQP_RTI can successfully solve and reduce the cost without any error. Any suggestion on the usage of SQP_RTI?
(I did not set rti_phase manually, which according to the Problem Formulation should be both for the RTI routine)