Optimization reaches minimal step length about hpipm HOT 2 CLOSED

Minimum step length reached may mean that the problem is unfeasible, or that the problem is feasible but somehow "hard" and the solver could not converge.

The Mehrotra's type interior point algorithm implemented does not give a certificate of infeasibility, so you can try to solve it with another solver (with e.g. an active set) to check which is the case.

Alternatively, you can check the stat field and the residuals at each iteration. If the problem is feasible you typically see that the solver is converging until at some point it starts diverging again: the final iterations of an interior point require the solution of ill-conditioned linear systems, and at that point the accuracy of the double precision computations may not enough. Switching to the robust mode (https://github.com/giaf/hpipm/blob/master/examples/c/example_d_ocp_qp.c#L137) may help, or the exit tolerances (https://github.com/giaf/hpipm/blob/master/examples/c/example_d_ocp_qp.c#L142 and following) should be reduced.

from hpipm.

I has a similar problem when solving an MPC problem. The Hpipm solver returns code 2. Strange to say, it would return code Sucess when removing the first general constraint. But, the MPC problem meet the first constraint. In addition, the MPC problem can be solved by OSQP. Switching to the robust mode and reducing the exit tolerances do not work.

from hpipm.