ROADMAP.md

Roadmap

• Test 32 bit
• Test OOQP
• More flexible solution sensitivities?
• get_optimal_value_hessian() - at least for HPIPM and exact Hessians
• remove old layer in Matlab interface

core

• propagate cost in integrator for NLS+IRK
  • or: add support for quadrature state, separate dimension in integrator and OCP solver
• faster workspace memory casting

ocp_nlp

• partial tightening
• RTI implementation similar to ACADO
• support cost on z for external, NLS

sim

• GNSF Hessians
• propagate cost in integrator for CONL+IRK
• time in integrator + time dependent model functions

DONE

• closed loop example MPC + MHE
• Templates: avoid global memory
• add support for manual model functions -- partly done: external cost and discrete dynamics

sim

• collocation integrators Radau
  • NOTE: currently always Gauss(-Legendre) Butcher tables
    • A-stable, but not L-stable
    • order is 2 * num_stages
  • implement also Radau IIA collocation methods
    • L-stable
    • order is 2 * num_stages - 1

C

• split ocp solve into prepare and feedback

build

• cmake: add openmp parallelization

matlab interface

• detect dimensions
• detect slack dimensions
• structure detections for constraints
• getting started folder
• add Mex templating support for: ( in prioritized order )
  • nonlinear least-squares
  • Vz
  • exact Hessian
  • external cost
  • GNSF
  • discrete dynamics
• separate acados_ocp() into generating the C object and setting the numerical data
• support nonuniform grids
• OCP with DAEs

Python interface

• exact hessian
• regularization
• discrete dynamics