Prompt guided object picking as a skiros2 skill.
This repository contains an implementation of Contact Graspnet and Grounded SAM in ROS1, as part of the Master's Thesis "Grasp Estimation for intelligent collaborative robot" in LTH, Lund University. This skill uses the Grounded SAM model to detect the object and executes the picking task using the Contact_Graspnet model and Moveit package.
- Tested on ROS Noetic, might work with other ROS distributions.
After installing the required packages above, follow these steps:
-
Clone this repository:
cd ~/catkin_ws/src git clone https://github.com/HashimHS/graspnet_skill.git
-
Add the skills to skiros2 by modifying the skill_mgr.launch file:
<arg name="libraries_list" value="[graspnet_skill]"/>
<arg name="skill_list" value="[detect_graspnet, move_graspnet]" />
You can also refer to this (guide) on how to add your skills to skiros2.
- Change the frame names in manipulation.py and graspnet.py according to your robot configuration. graspnet.py:
CAMERA_FRAME = 'realsense_rgb_optical_frame'
BASE_LINK = 'ur5e_base_link'
END_EFFECTOR_LINK = 'ur5e_ee_graspnet'
manipulation.py:
BASE_LINK = 'ur5e_base_link'
END_EFFECTOR_LINK = 'ur5e_ee_graspnet'
MOVE_GROUP = 'manipulator'
- Build the ROS workspace:
# Navigate to the root of your ROS workspace cd ~/catkin_ws # Build the workspace catkin build
Launch skiros2 gui using the command:
roslaunch skiros2 skiros.launch
This ROS package is made possible by: