Debugging input constraint violation

[avikde]

Related to #36

• simplify specification of experiment parameters using dicts
• reduce eps_prim_inf instead of eps

[avikde]

Pretty serious problems: getting

• solved inaccurate
• max iter reached

https://osqp.org/docs/interfaces/status_values.html

Don't really know how to solve this problem. :(

[avikde]

Maybe fixing #38 first might help make the trajectories more stable?

[avikde]

seems like the somersault and hover also doesn't work on master

[avikde]

changing eps_prim_inf / dual seem to have no effect on the input constraint violations.

[avikde]

• use u - Ax to check the violation
• print out the violation, and check the amount of violation. It must be less than eps_prim_inf

Some test results

• assert (res.x[-self.N*self.m.nu:-(self.N-1)*self.m.nu] == (self.A @ res.x)[-self.N*self.m.nu:-(self.N-1)*self.m.nu]).all() does not trigger; so the A setup is correct pertaining to limits on u
• np.amin(self.u - Ax), np.amin(Ax - self.l) are -0.00022063885292222807 -0.00159435525542619 (where u, l are the vectors of the same name provided to OSQP, Ax = A @ x), and the osqp settings are polish=False, scaling=0, eps_rel=1e-4, kdamping=0, eps_prim_inf=1e-6, eps_dual_inf=1e-6

[avikde]

Created this issue: osqp/osqp#125

Code to save the arrays at the suspicious time in mpc.py

				np.savez('arrays.npz', q=q, l=self.l, u=self.u)
				sparse.save_npz('P.npz', self.P)
				sparse.save_npz('A.npz', self.A)

[avikde]

Ok "solved inaccurate" basically means the solution is close but not quite there within the max number of iterations, so the problem can be removed by increasing max_iter. Now the problem is "primal infeasible" again. From inspecting inf_cert, it looks like u0 is infeasible? What does that mean?

TODO look at that row of A