Home

Welcome to the shared_autonomy_perception wiki!

This repo provides a framework designed for testing shared autonomy for perception in the context of manipulation. Thus, in addition to an implementation of Grabcut3D and human-machine interfaces, it includes a state machine that performs a simple table-clearing task (using MoveIt! for the manipulation parts). It also has the tools required for comparing fully autonomous ORK cluster-based segmentation with grabcut's human-assisted result.

The following diagram provides an overview of how the different pieces fit together.

Setup:

(assuming a PR2, but should be easily configurable to any robot controlled by MoveIt! with a kinect for perception)

• Install MoveIt!: `sudo apt-get install ros-hydro-moveit*
• Install PR2 specific packages: `sudo apt-get install ros-hydro-pr2*
• The rest assumes that your hydro workspace is ~/catkin_hydro
• Download the shared_autonomy_perception repo (master branch): cd ~/catkin_hydro/src; git clone https://github.com/SharedAutonomyToolkit/shared_autonomy_perception.git
• Download the cluster_grasp_planner (extracted from arm_navigation, released as a single package), (hydro-devel branch): cd ~/catkin_hydro/src; git clone https://github.com/bosch-ros-pkg/cluster_grasp_planner.git; cd cluster_grasp_planner; git checkout hydro-devel
• Install ORK, following the directions here

Running the demo

• (on robot): roslaunch shared_autonomy_launch robot.launch This is a slightly modified version of the /etc/ros/robot.launch file that starts the head-mounted kinect, but not any of the other cameras or the EKF localization.
• roslaunch pr2_moveit_config move_group.launch
• roslaunch pr2_moveit_config moveit_rviz.launch
• roslaunch shared_autonomy_launch clear_table_unified.launch use_grabcut:=true use_im:=false This launches the rest of the nodes required for clearing the table. Use the command line arguments to switch between shared autonomy using grabcut (use_grabcut:=true) and fully autonomous operation using ORK (use_grabcut:=false). You can also switch between using the Interactive Marker interface to grabcut (use_im:=false) and the JavaScript one (use_im:=true). If using Interactive Markers, be sure to add them to your rviz display. If using JavaScript, open this file in your browser: shared_autonomy_perception/js_hmi/pages/interactive_segmentation_interfaces.html

Other Notes

For information on creating and using logs, see the assemble_kinect package.

Several of the packages are obsolete:

• augmented_object_selection
• bosch_object_segmentation
• image_segmentation_demo

Others just contain example code:

• random_snippets
• webtools