rail_agile

Modified version of agile_grasp to expose functionality with ROS action servers using the message types defined in rail_grasp_calculation. The majority of the original agile_grasp package is included here, but this readme focuses on the changes made and how to run the AGILE pipeline with the new action server. For more information on the original package, see the AGILE readme.

find_grasps node

This node exposes action servers for executing either just antipodal grasp sampling (without classification), or the full AGILE pipeline (with classification). Relevant parameters, action servers, topics, and services are as follows:

• Action Servers
  • /rail_agile/sample_grasps(rail_grasp_calculation_msgs/SampleGraspsAction)
           Sample grasps over a point cloud within a given workspace. Requires a point cloud and workspace as input. Returns a list of antipodal grasps, unclustered and not passed through AGILE's SVM classifier. This will result in a large list of potential antipodal grasps.
  • /rail_agile/sample_classify_grasps(rail_grasp_calculation_msgs/SampleGraspsAction)
           Sample grasps over a point cloud within a given workspace, using the full AGILE pipeline. Requires a point cloud and workspace as input. Returns a list of antipodal grasps, clustered and passed through AGILE's SVM classifier. This will result in a small list of grasps that are more likely to be antipodal.
• Parameters
      See the AGILE documentation for a description of all parameters. They are unchanged here, but are important to set appropriately for best performance.

Installation

Clone the package into your catkin workspace and build it as follows:

cd (your catkin workspace)/src
git clone https://github.com/GT-RAIL/rail_agile.git
cd ..
catkin_make

Usage

Run the find_grasps node with the following, optionally setting parameters with rosrun or in a launch file:

rosrun rail_agile find_grasps

Make sure that all parameters to describe your robot's hand are set correctly, and set the num_samples parameter appropriately for the size of your point cloud.

To find grasps, create an action client that connects to either the sample_grasps action server or the sample_classify_grasps action server depending on the number and quality of grasps you would like. Pass in a point cloud that you want to search for grasps from, and a workspace that constrains the area within the point cloud to search for grasps in. The result will contain an unordered list of all of the grasp poses found by AGILE.