Skip to content

Latest commit

 

History

History
102 lines (77 loc) · 3.88 KB

README.md

File metadata and controls

102 lines (77 loc) · 3.88 KB
Raw

DOI

robot_hand_generator_MLS: evolutionary algorithm edition

A tool used to generate an optimal two-fingered gripper design for grasping an object of a certain width. Achieves this by using an evolutionary algorithm to find the design, and can run three different tests whose results can be compared. This tool can run three different tests and generates the best two-fingered gripper as found by each of those tests. The tests are:

Angle

The number of reachable angles around a center point

alt text

Straight (default)

Number of point pairs separated horizontally by width

alt text

Area

Number of points within the overlapping workspaces of both fingers

The test type can be specified in the params.py file. After testing at a coarse precision, the program takes the top few hands and tests them at a finer precision. To test each hand, the program generates the reachable area outline of each finger via the Denavit-Hartenberg parameters. It uses the Shapely library to determine if the points it is testing the reachability of fall within the outlines.

Parameters are in src/params.py and may be adjusted.

Requirements:

  • Linux OS or WSL with ubuntu

Install and Setup:

  1. Run the following commands:

    First, update your machine:

    sudo apt update
sudo apt upgrade

    Next, copy the following lines into the terminal:

    cd ~/

    git clone https://github.com/lsaltz/robot_hand_generator_MLS.git

    cd ~/robot_hand_generator_MLS/

    chmod u+x setup.sh

    ./setup.sh
    • This will set up the environment.

  2. After the environment has finished setting up (this will take some time), run the evolutionary algorithm, open:

    cd ~/robot_hand_generator_MLS/src/

    And run:

    python3 ea.py

    and specify for how many generations you would like to run it for. The more you enter, the longer it will take. For example, 5000 generations on my desktop takes 2-3 days.

output Folder:

After ea.py finishes running, it will send the data it gathered from each test on the top few grippers to two different results files located in ~/robot_hand_generator_MLS/output and ~/robot_hand_generator_MLS/points. In this folder you can find:

  • charts of the reachable space of the gripper when run through the test set in params
  • figure depicting the changes in maximum fitness score per generation
  • figures comparing segment lengths to fitness score
  • palm width to fitness score figure
  • brief summary of the results as a text file.

Please see Results.md for the current results.

Further Instructions:

If you wish to visualize the testing, run:

python3 visualization.py

to view the test set in params.py in simulation.

To run further tests on the winners of a single test, change the desired test in params.py and run:

python3 run_all_tests.py

Additional Notes:

Requires numpy 1.25.0, you may have to downgrade your numpy version. If you wish to run this software more than once, clear your outputs, hand_archive, and points folders. If you run into issues with MESA and OpenGL, add the following to ~/.bashrc:

export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6

This project was originally created by Josh Campbell, and has been adjusted by me. The original repository can be found at: https://github.com/OSUrobotics/robot_hand_generator

Advised by Kyle DuFrene, Keegan Knave, and Dr. Cindy Grimm.