ERK default precision?

byfra · February 20, 2020, 10:00pm

Hi !

I’m currently benchmarking ACADOS and IPOPT on simple OC problems. In order to operate a fair comparison between the 2, I was wondering what is the default precision of the ERK integrator implemented in ACADOS ?
In Casadi, it’s 20 finite elements, but it is possible to modify it.
And secondly, is it/will it be possible to modify it via the python interface ?

Thanks a lot !

Fb

FreyJo · February 21, 2020, 9:49am

Hi,

you can change the number of stages, Also you can change the number These options are in the AcadosOcp

   """ class containing the description 
"""
 def __init__(self):
     self.__qp_solver 
    self.__hessian_approx   = 'GAUSS_NEWTON' 
    self.__integrator_type  = 'ERK' 
    self.__tf               = None 
    self.__nlp_solver_type  = 'SQP_RTI' 
    self.__nlp_solver_step_length = 1.0 
    self.__sim_method_num_stages 
    self.__sim_method_num_steps   = 1 
    self.__sim_method_newton_iter = 3 
    self.__qp_solver_tol_stat = None 
    self.__qp_solver_tol_eq   = None 
    self.__qp_solver_tol_ineq = None 
    self.__qp_solver_tol_comp = None 
    self.__qp_solver_iter_max = 50 
    self.__qp_solver_cond_N = None 
    self.__nlp_solver_tol_stat = 1e-6 
    self.__nlp_solver_tol_eq   = 1e-6



i.e. size of the Butcher table for ERK to 1, 2 and 4.
 of steps, which is probably what you call “finite elements” to any integer.
 Options

/acados/blob/c63e3ebb437cf1c8fe0b62f37ea22af81f821a5e/interfaces/acados_template/acados_template/AcadosOcp.py#L1548" target="_blank">github.com os/acados/blob/c63e3ebb437cf1c8fe0b62f37ea22af81f821a5e/interfaces/acados_template/acados_template/AcadosOcp.py#L1548" target="_blank">acados/acados/blob/c63e3ebb437cf1c8fe0b62f37ea22af81f821a5e/interfaces/acados_template/acados_template/AcadosOcp.py#L1548 class="start lines" start="1538" style="counter-reset: li-counter 1537 ;"> of the solver options = 'PARTIAL_CONDENSING_HPIPM' # qp solver to be used in the NLP solver # hessian approximation # integrator type # prediction horizon # NLP solver # fixed Newton step length = 4 # number of stages in the integrator # number of steps in the integrator # number of Newton iterations in simulation method # QP solver stationarity tolerance # QP solver equality tolerance # QP solver inequality # QP solver complementarity # QP solver max iter # QP solver: new horizon after partial condensing # NLP solver stationarity tolerance # NLP solver equality tolerance

They cannot be modified after the ocp solver is created though.

Best,
Jonathan

Sijie · March 14, 2023, 8:34am

Hi, I have a question about how does number of steps in the integrator work? If I take integral step of 0.1s, and take the sim_method_num_stages = 4. Does that means the step of RK4 is 0.025s?

Tkans a lot.

FreyJo · March 14, 2023, 9:29am

yes, that is correct!