Python interface: possibility to change predicting horizon after solver is created

zhiyhu · September 15, 2023, 2:40pm

Hi!

I am implementing an MPC using the Python interface for acados. Originally the MPC has a horizon time tf = 1 and predicting horizon N = 20.

However, the MPC also requires shrinking the horizon if N = 20 is not feasible, so I would like to change tf and N after the AcadosOcpSolver object is created, during simulation.

I tried directly modifying the object attributes self.ocp.solver_options.tf and self.ocp.dims.N during simulation but they don’t seem to take effect.

May I ask if it is possible to modify these two or equivalent parameters online? If not possible, how can I recompile the solver object so the changes can take effect? Thanks a lot!!!

FreyJo · September 20, 2023, 7:37am

Hi,

you can use the function set_new_time_steps.

github.com

acados/acados/blob/9622cfdebdb06a91a4a65bf4fe42adf2a8f16152/interfaces/acados_template/acados_template/acados_ocp_solver.py#L1106


      
              """
              Sets current iterate to all zeros.
              """
              getattr(self.shared_lib, f"{self.model_name}_acados_reset").argtypes = [c_void_p, c_int]
              getattr(self.shared_lib, f"{self.model_name}_acados_reset").restype = c_int
              getattr(self.shared_lib, f"{self.model_name}_acados_reset")(self.capsule, reset_qp_solver_mem)
          
              return
          
          
          def set_new_time_steps(self, new_time_steps):
              """
              Set new time steps.
              Recreates the solver if N changes.
          
                  :param new_time_steps: 1 dimensional np array of new time steps for the solver
          
                  .. note:: This allows for different use-cases: either set a new size of time_steps or a new distribution of
                            the shooting nodes without changing the number, e.g., to reach a different final time. Both cases
                            do not require a new code export and compilation.
              """

zhiyhu · September 22, 2023, 11:39am

Great, thanks a lot!