Awesome

grasp_selection

• Author: Andreas ten Pas (atp@ccs.neu.edu)
• Version: 1.0.0
• Author's website: http://www.ccs.neu.edu/home/atp/

1) Overview

This package selects a subset of grasps from the grasps produced by the agile_grasp package. The grasps are selected based on three different scoring functions that take into account:

1. how easy it is to reach the grasp with the robot arm,
2. how easy it is to perform the grasp with the robot hand, and
3. how far the robot arm has to travel through the robot's workspace.

This package consists of one ROS node: select_grasps.

2) Requirements

1. ROS Hydro or ROS Indigo
2. agile_grasp package
3. MoveIt for providing Inverse Kinematics solutions

3) Installation

From Source, ROS Indigo

1. Open a terminal
2. Navigate to the src folder of your ROS workspace: $ cd location_of_workspace/src
3. Clone the repository: $ git clone https://github.com/atenpas/grasp_selection.git
4. Navigate back to the root of your ROS workspace: $ cd ..
5. Recompile your ROS workspace: $ catkin_make

From Source, ROS Hydro

Same as 3.1, except for Step (3): $ git clone https://github.com/atenpas/grasp_selection.git -b hydro

4) Usage

This package requires the MoveIt /compute_ik ROS service to calculate Inverse Kinematics. The following two steps explain how to start this service when using a Baxter robot. First, start the joint trajectory action server:

rosrun baxter_interface joint_trajectory_action_server.py

Next, launch rviz with moveit:

roslaunch baxter_moveit_config demo_baxter.launch

The next step is to launch an agile_grasp node (check the link for further instructions):

roslaunch agile_grasp baxter_grasps.launch

Now, the grasp_selection node can be launched:

roslaunch grasp_selection select_grasps.launch

The node works by first testing each grasp for reachability. All the remaining grasps are then scored according to the three scoring functions listed above. The node finally selects the k top scoring grasps.

The grasp selection node provides the selected grasps through a ROS service (see srv/SelectGrasps.srv). The grasping demo mentioned below contains example code for accessing this service.

5) Grasping Demo

The package comes with two grasping demos, one that uses OpenRAVE, and one that uses MoveIt. MoveIt is the simpler way of running the grasping demo.

MoveIt

rosrun grasp_selection moveit_grasping.py

OpenRAVE

rosrun grasp_selection openrave_grasping.py

Remark: To run the grasping demo, you need to follow the steps listed in (4) Usage. In addition, a ROS node needs to be running that provides point clouds in the same frame in which arm motions are planned. The point clouds produced by that node need to be the input to the agile_grasp and grasp_selection packages.

6) Parameters

The parameters in the ROS launch file select_grasps.launch are described below.

General

• cloud_topic: the ROS topic at which the input point cloud for agile_grasp is published (used for collision avoidance)
• grasps_topic: the ROS topic where agile_grasp publishes the grasps
• joint_states_topic: the ROS topic for joint states
• marker_lifetime: the lifetime of visual markers in Rviz
• use_scoring: whether the grasps are scored

Reachability

• workspace: the dimensions of the robot's workspace
• min_aperture: the minimum aperture of the robot hand
• max_aperture: the maximum aperture of the robot hand
• axis_order: the ordering of the axes of the robot hande frame (e.g., [2, 0, 1] for the Baxter research robot)
• planning_frame: the planning frame of the robot
• hand_offset: the distance between the finger tip and the base of the robot hand
• num_additional_grasps: the number of additional grasps that is produced
• arm_link: name of the robot arm end effector link (required by MoveIt)
• move_group: name of the robot arm "move group" (required by MoveIt)
• max_colliding_points: the maximum number of points that is allowed to be in collision
• JS_first_joint_index: the index of the first arm joint on the joint_states ROS topic
• JS_last_joint_index: the index of the last arm joint on the joint_states ROS topic
• IK_first_joint_index: the index of the first arm joint in the IK solver's solution
• IK_last_joint_index: the index of the last arm joint in the IK solver's solution
• planning_library: which motion planning library is used for solving IK (0: MoveIt, 1: OpenRAVE)
• prints: whether additional information is printed during reachability tests

Notice: When using OpenRAVE as the planning_library, the ikfast solver ROS service contained in this package needs to be started:

rosrun grasp_selection ikfast_service.py

This service provides IK solutions using OpenRAVE. They are used for reachability and scoring in the grasp selection process.

Scoring

• urdf: the location of the URDF file
• num_selected: the number of selected grasps