SWIG 将 C++ 包装成 Python

Question

I wrote a C++ module to compute the forward and inverse kinematics for my SCARA robot arm, and I want to wrap this module into Python so that I can use it in another application. For the conversion, I chose SWIG, but it is pretty hard to write the interface file correctly.

In my header of C++ module, I have

namespace ARM_KINEMATICS{
    /*
       Forward kinematics
       [in] const double *q: joints' value
       [out] double *T: placeholder to put homogeneous transformation matrix, with length 16
     */
    void forward(const double *q, double *T);
    /*
       Inverse kinematics
       [in] const double *T: target homogenous transformation matrix with length 16
       [out] double q_sols: placeholder for solutions, with length 2*4
       [out] int: number of solutions
     */
    int inverse(const double *T, double *q_sols);
}

The expected behavior in Python would be like

> import arm_kinematics
> T = [0.0] * 16
> arm_kinematics.forward([0.0, 0.0, 0.0, 0.0], T)
> T
  [1.0, 0.0, 0.0, 0.6, 0.0, -1.0, 0.0, 0.0, 0.0, 0.0, -1.0, 0.139, 0.0, 0.0, 0.0, 1.0]
> q_sols = [0.0] * 8
> num_sols = arm_kinematics.inverse(T, q_sols)
> q_sols[:num_sols*4]
  [0.0, 0.0, 0.0, 0.0]

How can I write the interface file so that it can convert Python list into C++ array and vice verse? And is it possible to make Python pass by reference ? I found

typemap

can handle functions with known array size, for example int foo(int arr[4]) , but if now the array is passed by the pointer and we don't know its size, how should I revise the interface file?

Thank in advance!

Answer 1

The easiest way using SWIG is to use numpy.i

Change your C++ interface to this

namespace ARM_KINEMATICS{
  /*
    Forward kinematics
    [in] const double *q: joints' value
    [out] double *T: placeholder to put homogeneous transformation matrix, with length 16
  */
  void forward(const double *q, const int nq, double **T, int* nT);
  /*
    Inverse kinematics
    [in] const double *T: target homogenous transformation matrix with length 16
    [out] double q_sols: placeholder for solutions, with length 2*4
    [out] int: number of solutions
  */
  int inverse(const double *T, const int nT, double **q_sols, int* nq_sols);
}

A working interface for SWIG is then

%module robot
%{
  #define SWIG_FILE_WITH_INIT
  #include "robot.h"
%}

%include "numpy.i"

%init
%{
  import_array();
%}

%apply (double* IN_ARRAY1, int DIM1) {(const double *q, const int nq)}

%apply (double* IN_ARRAY1, int DIM1) {(const double *T, const int nT)}

%apply (double** ARGOUTVIEWM_ARRAY1, int* DIM1) \
{(double** T, int* nT)}

%apply (double** ARGOUTVIEWM_ARRAY1, int* DIM1) \
{(double** q_sols, int* nq_sols)}


%include "robot.h"

If you look into numpy.i , you will see that you can input/output matrices as well.

The typemaps are defined in numpy.i , which is shipped together with NumPy