robotsorcerer / torobo

One Hell of a Control Lyapunov Function Solver for Nonlinear Control Tasks

Header Info

author = "Olalekan Ogunmolu"

copyright = "2018, One Hell of a Lyapunov Solver"

credits = "Rachel Thompson (MIT), Jethro Tan (PFN), Ayaka Kume (PFN)"

license = "MIT"

maintainer = "Olalekan Ogunmolu"

email = "[e-mail]"

status = "Testing"

Introduction

This codebase represents a wip for our proposed learning stable control laws for nonlinear complex robot trajectories.

Packages Organization

• Torobo_ik: Contains the ik solver and service calls' source code for converting the joint positions of the robot to cartesian workspace. This leverages on the TRAC_IK_SOLVER and the KDL_LIBRARY.

• Torobo Control: Details the meta-package for the whole packages that we use in this project.

• LyapunovLearner: Basic implementation of Khansari-Zadeh's Stable Estimator of Dynamical Systems paper and Learning Control Lyapunov Functions for Nonlinear Robot Trajectories.

• LWPR: This is an implementation of Stefan Schaal and Sethu Vijaykumar's Locally Weighted Projected Regression from their 2003 ICML paper. This is the test Matlab source code for

  • 1D data,

  • 2D data, and

  • nD data

  The original matlab lwpr code is located here.

• LWPR Translations: These are my translations for the example that the original authors provided in the matlab code.

  • This can be tested with python test_lwpr_1D.py.

• LWPR 3rd party python source: This contains the source code for the python/c/mex source code by a 3rd party developer. I earlier tried to install this on a linux system as the readme file directs but I found that after running configure, make was giving weird errors. Would appreciate if you could look into this.

Running

• Bring up the robot

  roslaunch toroboarm_seven_bringup bringup_real.launch

  roslaunch torobo_ik torobo.launch

• The above two commands can be merged with

  roslaunch torobo_ik torobo.launch bringup:=true

  In addition, one can turn off the stdout printouts by appending the arg disp:=true to the torobo.launch file e.g.,

  roslaunch torobo_ik torobo.launch bringup:=true disp:=false

• Launch the Lyapunov Learner Executor

  In a separate terminal, launch the robot executor.

  cd /path_to_root_folder/lyapunovlearner/scripts python main.py

Update: Aug 16, 2018. Tokyo

• CSV raw files from Torobo now in lyapunovlearner/scripts/data/aug_16 folder

• Note that these files represent each motion for the same sinusoidal stuff for the robot. The data was collected in joint space.

• They have been augmented with the target position in the files to make the data uniform across all demos. See the scripts in torobo_ik/src/joints_pub.py

• The joint space data collected was converted to cartesian workspace coordinates using the KDL library. This is saved in lyapunovlearner/scripts/data/cart_pos_new.csv

• We use the cartesian coordinates of the data collected in teaching the robot to repeat the motion.