Open dlcole3 opened 2 years ago
not only A
and B
, we should allow everything to be time-varying Q,R,S,K
. Also, we should consider linear offset terms and allow them to be time-varying. E.g., consider q,r,d
satisfing
stage cost: x' * (0.5 * Q*x - q) + u' * (0.5 * R*u - r) + x' * S *u
dynamics: x[t+1] = Ax[t] + Bu[t] + d[t]
Actually, supporting time-varying q,r,d
should not be too difficult, but supporting time-varying matrices can be non-trivial. Would be good to first start with time-varying vectors
Currently,
DynamicNLPModels.jl
only takes an input of a singleA
orB
matrix, but these can vary with time steps. Need to add support for handling a set of $N$A
orB
matrices (e.g., $A_i \forall i = 0, 1, ..., N-1$).