GMP - Generalizable Movement Primitives Library.

GMP is library that provides an interface for DMP (Dynamic Movement Primitives). It is based on the DMP formulation presented in this paper

---

Main Features

• Encoding and generalization of a kinematic behaviour.
• Online adaptation to via-points.
• Off-line optimization under kinematic constraints.
• Onl-ine MPC-like optimization under kinematic constraints.

---

Supported Interfaces

• matlab
• c++

Both interfaces are implemented in a similar way employing the same classes and functions (except for some small exceptions due to the specific structure of each language). The matlab libraries are located in src/lib/matlab and the c++ libraries in src/lib/c++.

---

Dependencies for C++ interface

• catkin

• armadillo It's best to have the latest version of armadillo installed. Older versions might cause strange runtime errors (that are not related to armadillo in any obvious way).

• OSQP library Binaries are already included in the osqp_lib folder which is ros package wrapper for the libary. To update the library to the latest version, run:

    ./install/osqp_install.sh

---

Examples

Examples are provided in the package src/gmp_test both in matlab and c++.

Setup

To run the matlab code, first setup your matlab search path. To do so, open a matlab, browse to src\gmp_test\src and from the matlab command prompt type:

  add_libraries_to_matlab_path

To run the c++ code, you have to build it first. To do so, open a terminal, browse to the repository head and type:

 ./build.sh

Run

• Encoding and reproduction of a Cartesian position trajectroy:

  matlab: from matlab, browse to src\gmp_test\src\simulation and run:

    nDoF_gmp_test

  c++: open a terminal, browse to the repository head and run:

    source devel/setup.bash
    roslaunch gmp_test nDoF_GMP_test.launch

  To plot the results, from matlab browse to src\gmp_test\src\simulation and run:

    nDoF_gmp_plot

  You can also alter the simulation parameters by changing the ros parameters from src\gmp_test\launch\nDoF_GMP_test.launch

• Encoding and reproduction of a Cartesian orientation trajectroy:

  matlab: browse from matlab to src\gmp_test\src\simulation and run:

    orient_gmp_test

  c++: open a terminal, browse to the repository head and run:

    source devel/setup.bash
    roslaunch gmp_test orient_GMP_test.launch

  To plot the results, browse from matlab to src\gmp_test\src\simulation and run:

    orient_gmp_plot

  You can also alter the simulation parameters by changing the ros parameters from src\gmp_test\launch\orient_GMP_test.launch

• Adding via-points:

• Enforcing constraints offline:

• Enforcing constraints online:

---

Miscellaneous

• IO functionalities

  The package has additionally IO functionalities for reading/writing data from/to binary files, provided by the class FileIO. Each variable has a name-identifier assigned to it, which is used to write or read the variable. In general, it works like this for writing:

  % data to write
  A = rand(4,5);
  v = rand(1,3);
  str = 'Hello World!';
  gmp = GMP(6, 30); % DMP with 6 DoFs and 30 kernels per DoF
  gmp_o = GMPo(40); % orientation DMP with 40 kernels
  
  % write to file
  fid = FileIO('my_data.bin', FileIO.out);
  fid.write('my_vector', v);
  fid.write('my_message', str);
  fid.write('some_matrix', A);
  gmp_.write(gmp, fid, 'my_gmp'); % use dedicated function for writing GMP
  gmp_.write(gmp_o, fid, 'o_gmp');

  and to read:

  % read from file (in any order you like)
  fid = FileIO('my_data.bin', FileIO.in);
  
  gmp = GMP(); % create empty object
  gmp_.read(gmp, fid, 'my_gmp'); % use dedicated function for reading GMP
  
  A = fid.read('some_matrix');
  
  gmp_o = GMPo(); % create empty object
  gmp_.read(gmp_o, fid, 'o_gmp');
  
  v = fid.read('my_vector');
  str = fid.read('my_message');

  To see more details on how it works, check the examples io_example.m and gmp_io_example.m in the folder src\gmp_test\src\simulation. The API is almost identical for c++ and compatible with the matlab API. Therefore you can read from c++, data written in matlab and vice versa.

---

Citing GMP

If you are using GMP for your work, you are encouraged to

• Cite the related papers:

  @INPROCEEDINGS{9562059,  
  author={Sidiropoulos, Antonis and Doulgeri, Zoe},  
  booktitle={2021 IEEE International Conference on Robotics and Automation (ICRA)},   
  title={A Reversible Dynamic Movement Primitive formulation},   
  year={2021},  
  volume={},  
  number={},  
  pages={3147-3153},  
  doi={10.1109/ICRA48506.2021.9562059}}
  

• Put a star on this repository.