Chess for Robots

This is the n-th generation chess code, which was originally developed for AAAI 2011.

Installation

sudo apt-get install gnuchess gnuchess-book python3-pexpect
sudo apt-get install festlex-cmu
sudo apt-get install ros-noetic-moveit-full ros-noetic-moveit-python

cd <catkin_ws>/src
wstool init
wstool merge https://raw.github.com/mikeferguson/chessbox/master/maxwell.rosinstall
wstool update
cd ..
catkin_make

Setup for festival

cd /usr/share/festival/voices/english
sudo wget -c http://www.speech.cs.cmu.edu/cmu_arctic/packed/cmu_us_awb_arctic-0.95-release.tar.bz2
sudo tar jxf cmu_us_awb_arctic-0.95-release.tar.bz2 
sudo ln -s cmu_us_awb_arctic cmu_us_awb_arctic_clunits

Running on Maxwell -- pretty much only I can do this...

roslaunch maxwell_defs bringup_maxwell.launch
roslaunch maxwell_defs head_camera.launch
roslaunch maxwell_moveit_config move_group.launch
roslaunch chess_player play.launch

Running without the robot

You can startup most of the pipeline without having a Maxwell:

roslaunch maxwell_defs fake_maxwell.launch
roslaunch maxwell_moveit_config move_group.launch
rosrun tf static_transform_publisher .15 .2286 .7366 -1.57 0 0 base_link chess_board 20

You probably want to view things. So "rosrun rviz rviz". Be sure to add a PlanningScene display, and set the Planning Scene topic to "move_group/monitored_planning_scene". You can then run through some tests. For instance, it's nice to tuck the arm (it will start up pointing straight out):

rosrun chess_player tuck_arm.py

The next test is to move all pawns forward 2 steps. You should be able to watch the chess board be created, a bunch of little blocks be inserted, and then all of the "pawns" be moved forward. Startup may take some time:

rosrun chess_player grasp_utilities.py

Finally, you might be able to run the full executive and have the robot move some pieces, but this does get broken from time to time:

rosrun chess_player chess_executive.py --sim

Running on UBR-1

rosrun tf static_transform_publisher .4 .2286 .7366 -1.57 0 0 base_link chess_board 20

Adapting to a new robot

Much of this code is robot-independent. Below I list the files known not to be (in other words, if the file isn't in this list, it is probably robot-independent).

chess_perception

The only robot-related aspect is that the head should be pointed at the board. Also, if your robot does not have base_link, you will need to set ~fixed_frame for the chess_perception_node. The other major issue is that the perception code currently is not entirely robust to pieces being improperly placed on the board.

chess_player/src/robot_defs.py

All of this file is potentially robot-specific.

chess_player/src/head_utilities.py

This is currently hard coded with angles for Maxwell. Ideally this would eventually turn into a point_head action, with a "search and then cache" function for "looking at the board".

chess_player/src/grasp_utilities.py

Something should be done with OFF_BOARD positions used for capture.

The functions for getGripperPosture, getGripperTranslation, and all of the grasp generation stuff (getY, getP, getGrasps, getPlaceLocations) are Maxwell-specific, although if your arm is pretty good at overhead grasps this may just work for you.

chess_player/nodes/tuck_arm.py

This just calls grasp_utilities.py. Obviously, this is currently hard-coded with the tuck positions of Maxwell. Ideally we would load the tuck position from the SRDF file to make this robot-independent.

chess_player/nodes/tilt_head.py

This calls head_utility.py stuff.