ds_motion_generator

---

[This README is outdated -- will clean it up soon!]

This package provides a nodified version of DS motion generators. Type of DS:

• Analytically parameterized DS for simple motions as in [1], such as
  • linear motions, point-to-point oscillatory motions and swiping as below

- quaternion linear motion [3]
- joint-space linear motion [4]

• Non-linear DS learned from demonstrations with the following approach:

• lpv-DS parametrization [2]. For testing, we provide a couple of pre-learned models of non-linear, non-monotic motions used in a variety of tasks in [3]; e.g. sink motion, via-point motion, CShape motion

• lags-DS parametrization [2]: See nadia-lags branch

• DS Modulation Strategies

  • See modulation branch
  • Human/Obstacle Collision Avoidance (ongoing work with Christopher Fourie)
  • DS State-Space Encapsulation for Task Bi-Simulation Guarantees (ongoing work with Felix Wang)

Installation

Do the following steps:

• In your catkin src directory clone the repository

$ git clone -b latest-mit https://github.com/nbfigueroa/ds_motion_generator.git

• wstool gets all other git repository dependencies, after the following steps you should see extra catkin packages in your src directory.

$  wstool init
$  wstool merge ds_motion_generator/dependencies.rosinstall 
$  wstool up 

• Query and installs all libraries and packages

$ rosdep install --from-paths . --ignore-src --rosdistro indigo 

Usage

DS Configuration and Launch File Specifications

• To test the analytical DS motions you can run any of the following launch file as the ones found in the launch/ folder. For the linear DS, you should fill in the DS configuration file, which is a .yaml file stored in config/analytic_DS folder, the default is set to -y direction of motion. You can then test the launch file as follows:

  $ roslaunch ds_motion_generator load_linearDS_motionGenerator.launch 
  

  Some imporant things to fill in the launch file are the following:

  • The name of the input topic (potentially a position signal)
  • The name of the output topic (potentially a desired velocity signal)
  • The name of the topic for the filtered output.

  The current topics are assigned assuming you are using the kuka-lwr-ros control interface and simulator (Gazebo). For the other analytical DS, the following launch files are provided:

  $ roslaunch ds_motion_generator load_ppOscDS_motionGenerator.launch viz_DS_path:=true
  

  The variable viz_DS_path defines if you want to visualize the integrated path of the DS.

  For the swipe motion the following launch file is provided:

  $ roslaunch ds_motion_generator load_swipeDS_motionGenerator.launch direction:=right viz_DS_path:=true
  

  The variable direction defines the direction of motion, which can be left/right.

• To test the learned DS via se-DS [2] parametrization, we provide the following launch file

  $ roslaunch ds_motion_generator load_seDS_motionGenerator.launch DS_name:=free_ws viz_DS_path:=true
  

  in which,apart from the input/output variables defined above, you must indicate the DS configuration file, which should contain the following information in YAML format:

  • K (number of guassian)
  • dim (the dimenstion input-output space)
  • Priors
  • Mu
  • Sigma
  • attractor
    Sometimes the parameters of Mu and Sigma might be too large; i.e. >1e6. In these cases you can scale them and add the scaling values to the yaml file:
  • Mu_scale
  • Sigma_scale

  This file is stored in the config/learned_DS/seDS folder. The input variable DS_name in the launch file is used to indicate the names of the pre-learned se-DS models, which are: push_down,Curve_go_right,Curve_go_left,free_ws.

  To learn/test your own se-DS model, you must generate this yaml file, which you can do by following the demo_learn_seDS.m script in the ds-opt package [3].

• To test the learned DS via lpv-DS [2] parametrization, we provide the following launch file

  $ roslaunch ds_motion_generator load_lpvDS_motionGenerator.launch DS_name:=sink-pqlf viz_DS_path:=true
  

  in which,apart from the input/output variables defined above, you must indicate the DS configuration file, which should contain the following information in YAML format:

  • K (number of guassian)
  • dim (the dimenstion input-output space)
  • Priors
  • Mu
  • Sigma
  • A
  • attractor

  This file is stored in the config/learned_DS/lpvDS folder. The input variable DS_name in the launch file is used to indicate the names of the pre-learned se-DS models, which are: sink-pqlf,via-point-pqlf,CShape-top-pqlf.

  To learn/test your own lpv-DS model, you must generate this yaml file, which you can do by following the demo_learn_lpvDS.m script in the ds-opt package [3].

---

Dynamic re-configuration of DS/filter parameters

Finally, once running the node for your desired DS, you can modify the filtering parameters and some of the DS parameters dynamically. The definition of the parameters that can be dynamically reconfigured is provided in the cfg/. Following an example of the parameters you can reconfigure:

• Wn : speed of the filter
• fil_dx_lim : the limit on the velocity of the filter. (Note this is not the limit of the real velocity).
• fil_ddx_lim : the limit on the accelarion of the filter. (Same note)
• target_{x,y,z} : the position of the attractor.
• scaling : simply a multiplier to scale the computed velocities based on the DS of choice.
• trimmit : simply a threshold to limit the computed velocities based on the DS of choice.

References

[1] Khoramshahi, M. and Billard, A. (2018) A Dynamical System Approach to Task-Adaptation in Physical Human-Robot Interaction. Autonomous Robots.

[2] Figueroa, N. and Billard, A. (2018) A Physically-Consistent Bayesian Non-Parametric Mixture Model for Dynamical System Learning. In Proceedings of the 2nd Conference on Robot Learning (CoRL).

This package was initially implemented by Mahdi Khoramshahi and has been extended and modified by Nadia Figueroa.

Contact: Nadia Figueroa (nadia.figueroafernandez AT epfl dot ch)

Acknowledgments This work was supported by the European Communitys Horizon 2020 Research and Innovation pro- gramme ICT-23-2014, grant agreement 644727-Cogimon and 643950-SecondHands.