kabzo
March 26, 2020, 6:47am
1
Hi,
is it possible to create discrete dynamics from python interface? The reason I would want discrete dynamics is that I would like to use variable timestep along the prediction horizon.
Thanks
FreyJo
March 27, 2020, 4:12pm
2
Hi @kabzo ,
I was implementing nonuniform discretizations for Python yesterday.https://github.com/FreyJo/acados/tree/nonuniform_templates/
#
# Copyright 2019 Gianluca Frison, Dimitris Kouzoupis, Robin Verschueren,
# Andrea Zanelli, Niels van Duijkeren, Jonathan Frey, Tommaso Sartor,
# Branimir Novoselnik, Rien Quirynen, Rezart Qelibari, Dang Doan,
# Jonas Koenemann, Yutao Chen, Tobias Schöls, Jonas Schlagenhauf, Moritz Diehl
#
# This file is part of acados.
#
# The 2-Clause BSD License
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
#
# 1. Redistributions of source code must retain the above copyright notice,
# this list of conditions and the following disclaimer.
#
# 2. Redistributions in binary form must reproduce the above copyright notice,
# this list of conditions and the following disclaimer in the documentation
# and/or other materials provided with the distribution.
#
show original
#
# Copyright 2019 Gianluca Frison, Dimitris Kouzoupis, Robin Verschueren,
# Andrea Zanelli, Niels van Duijkeren, Jonathan Frey, Tommaso Sartor,
# Branimir Novoselnik, Rien Quirynen, Rezart Qelibari, Dang Doan,
# Jonas Koenemann, Yutao Chen, Tobias Schöls, Jonas Schlagenhauf, Moritz Diehl
#
# This file is part of acados.
#
# The 2-Clause BSD License
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
#
# 1. Redistributions of source code must retain the above copyright notice,
# this list of conditions and the following disclaimer.
#
# 2. Redistributions in binary form must reproduce the above copyright notice,
# this list of conditions and the following disclaimer in the documentation
# and/or other materials provided with the distribution.
#
show original
There might still be some changes on how it is to be used, e.g. I am thinking if it is more convenient to pass shooting nodes or time steps.
Feel free to try or ask if something is not clear
Cheers,
kabzo
March 31, 2020, 1:14am
3
Thanks @FreyJo !!! I did try the examples and it was exactly what I wanted, thanks!
I tried your latest PR with my OCP, https://github.com/acados/acados/pull/573 , and it works well!
Thanks
1 Like
Hi Jonathan,
Thank you for the feature!
But one question. Is it possible to set nonuniform discretizations in runtime after a solver has been compiled?
Regards,
FreyJo
June 3, 2020, 11:45am
5
Hi Alex,
this is not possible at the moment.
Best,
Thank you for fast reply!
How is it difficult? If it does not touch on the core functionality, maybe I will be able to implement the feature?
Regards,
What has to be done in C, is this:
double time_steps[N];
{%- for j in range(end=dims.N) %}
time_steps[{{ j }}] = {{ solver_options.time_steps[j] }};
{%- endfor %}
for (int i = 0; i < N; i++)
{
ocp_nlp_in_set(nlp_config, nlp_dims, nlp_in, i, "Ts", &time_steps[i]);
ocp_nlp_cost_model_set(nlp_config, nlp_dims, nlp_in, i, "scaling", &time_steps[i]);
}
In the Python interface it would have to be added in the setter:
else:
out = np.ascontiguousarray(np.zeros((1,)), dtype=np.float64)
out_data = cast(out.ctypes.data, POINTER(c_double))
self.shared_lib.ocp_nlp_get.argtypes = [c_void_p, c_void_p, c_char_p, c_void_p]
self.shared_lib.ocp_nlp_get(self.nlp_config, self.nlp_solver, field, out_data)
return out
# Note: this function should not be used anymore, better use cost_set, constraints_set
def set(self, stage_, field_, value_):
cost_fields = ['y_ref', 'yref']
constraints_fields = ['lbx', 'ubx', 'lbu', 'ubu']
out_fields = ['x', 'u', 'pi', 'lam', 't']
# cast value_ to avoid conversion issues
value_ = value_.astype(float)
field = field_
field = field.encode('utf-8')
In Matlab/Octave, this is the setter function:
#include <stdio.h>
#include <string.h>
// acados
#include "acados_c/ocp_nlp_interface.h"
#include "acados_c/sim_interface.h"
// mex
#include "mex.h"
#include "mex_macros.h"
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
long long *ptr;
int acados_size;
mxArray *mex_field;
char fun_name[20] = "ocp_set";
char buffer [500]; // for error messages
/* RHS */
int min_nrhs = 5;
For the GNSF-IRK integrator it is assumed to be constant, I am not sure about other modules.
Since this might cause issues, I am still not sure if we want to have this in the master.
Cheers
1 Like
Hi @FreyJo , I am trying to run the examples discussed here and I am not able to find them. Please, could you say me where I can get them?
Thanks in advance
josu
