grasp-factory

Automatic Grasp Success/Failure Checker based on Gazebo (ROS1). Depending on user's inputs (e.g., list of planned grasps), the program spawns an object and a gripper (Gripkit-CR-Plus-L) models to check if each planned grasp succeeds or fails.

The project was done while Hojun Lee was working for Barton Research Group (BRG) at the University of Michigan.

Table of Contents

• Repository Structure
• Download Process
• Simulation
  • Gazebo Only
  • MoveIt Only
  • MoveIt and Gazebo
• Automated Grasping
  • How to Run
  • Grasp Representation
• ToDo Lists

---

Repository Structure

├── auto_grasp
│   └── src                           # Python source codes
├── gripkit_cr_plus_l_bb_description
│   ├── launch                        
│   ├── meshes                        # STL files
|   └── urdf                          # URDF description
├── gripkit_cr_plus_l_bb_moveit
│   ├── config                        # config files
│   └── launch                        # ROS-launch files
└── models
    ├── dict                          # grasp dictionaries
    └── obj_05                        # object folder

Download Process

Note This repository has been tested on ROS Noetic and Ubuntu 20.04. It also depends on numpy, scipy, and tqdm:

cd ~/catkin_ws/src
git clone https://github.com/kidpaul94/grasp-factory.git
cd grasp-factory/
pip3 install -r requirements.txt
cd ~/catkin_ws
catkin_make
source devel/setup.bash

Simulation

Gazebo Only:

Note This launch file is only for model visualization purpose. No ROS controllers will get initialized.

roslaunch gripkit_cr_plus_l_bb_description gazebo.launch

MoveIt Only:

roslaunch gripkit_cr_plus_l_bb_moveit_config demo.launch

MoveIt and Gazebo:

roslaunch gripkit_cr_plus_l_bb_moveit_config demo_gazebo.launch

Automated Grasping

How to Run:

Note simulation.py receives several different arguments. Run the --help command to see everything it receives.

cd auto_grasp/src
python3 simulation.py --help

Grasp Representation:

Note The representation is defined based on a coordinate system of each object 3D model.

The image above shows the grasping center & direction & approach vector format that we use to spawn an object at specific poses. In simulation.py, we utilize a list of this representation as a grasp dictionary associated with a specific object model and check success/failure of each planned grasp.

ToDo Lists

Model & Controller parameters tuning
Documentation