SkinLearn

SkinLearn is a multi-modal skin based Human-Robot Interaction learning, estimation and control system. This includes libraries for Intelligent Control and Estimation (ICE).

Branches:

master (Groovy, rosbuild) - deprecated
rbfnn (test with RBF activation functions) - incomplete
catkin-devel - started converting packages from rosmake to catkin
indigo-devel - main branch, work in progress

Working packages:

ice_msgs
ice_robot_controllers: contains the adaptNeuroController class
ice_experimenter: keyboard interface for running experiments
ice_sensors: code for tactile sensors
pr2_cart: real-time admittance and velocity controller for PR2 cart following behavior
uta_pr2_...: older packages that no longer work (and to be removed)

Workspace setup

Use the following steps to setup a ROS indigo workspace:

source /opt/ros/indigo/setup.bash
mkdir ~/ice_ws/src -p
cd ~/ice_ws/src
catkin_init_workspace
cd ..
catkin_make

echo "source ~/ice_ws/devel/setup.bash" >> ~/.bashrc
source ~/.bashrc

Install Dependencies

The following needs to be installed:

sudo apt-get install ros-indigo-pr2-desktop
sudo apt-get install ros-indigo-pr2-gripper-sensor*
sudo apt-get install ros-indigo-cmake-modules

The ice_sensors package requires the Serial Communication Library:

cd ~/ice_ws/src
git clone https://github.com/wjwwood/serial.git
cd ..
catkin_make

Install ICE Library

Download the repositories and build:

cd ~/ice_ws/src
git clone https://<USER>@bitbucket.org/<USER>/uta_pr2-dev2.git -b catkin-devel
git clone https://<USER>@bitbucket.org/nextgensystems/robot_utilities.git
cd ..
catkin_make

To import the workspace into Eclipse IDE, do this once:

cd ~/ice_ws
catkin_make --force-cmake -G"Eclipse CDT4 - Unix Makefiles"

The branch can be changed using the following command:

cd /path/to/repository
git checkout catkin-devel
git status

Troubleshooting

If the build fails due to the pr2 gripper sensor package, try compiling it from source:

cd ~/ice_ws/src
git clone https://github.com/PR2/pr2_gripper_sensor.git -b hydro-devel

Demo 1: Tactile Sensors

Make sure the sensor data can be read from the USB port:

sudo apt-get install gtkterm
gtkterm

and do a shift+ctrl+s. Choose the correct port parameters, i.e.

port: /dev/ttyUSB0
baud: 115200

If data is being published, close gtkterm (note: only one program can read from the port). Adjust the parameters in /ice_sensors/config/tactileBox.yaml and start the ROS driver/RVIZ using

roslaunch ice_sensors tactileBox.launch
roslaunch ice_sensors rviz.launch

Demo 2: Neuroadaptive controller

The most up-to-date controller is inside the adaptNeuroController class which can be launched with

roslaunch ice_robot_controllers adaptNeuroController.launch

To use the ICE experimenter:

roslaunch ice_experimenter ice_experimenter.launch NN:=true JT:=false
rosrun ice_experimenter ice_experimenter

Check the launch file and configuration files in ice_experimenter/cfg for more settings. Note: Run the keyboard interface on the PR2 to avoid timing offsets during data collection.

Running in Simulation

Add the following settings to bashrc:

export ROBOT=sim
export ROS_MASTER_URI=http://localhost:11311

Make sure the following parameters in ice_robot_controllers/config/pr2_controller.yaml are set to false:

useFTinput         : false
forceTorqueOn      : false
useFlexiForce      : false
accelerometerOn    : false

Start gazebo and launch the controller:

roslaunch pr2_gazebo pr2_empty_world.launch
roslaunch ice_robot_controllers adaptNeuroController.launch

Manually start the controller with the pr2_controller_manager

rosrun pr2_controller_manager pr2_controller_manager list
rosrun pr2_controller_manager pr2_controller_manager stop l_arm_controller
rosrun pr2_controller_manager pr2_controller_manager start pr2_adaptNeuroController

or use the keyboard utility:

rosrun ice_experimenter ice_experimenter

To manually apply a force:

rosservice call /gazebo/apply_body_wrench "body_name: 'l_gripper_l_finger_tip_link'
reference_frame: 'l_gripper_l_finger_tip_link'
reference_point: {x: 0.0, y: 0.0, z: 0.0}
wrench:
  force: {x: 1.0, y: 4.0, z: 0.0}
  torque: {x: 0.0, y: 0.0, z: 0.0}
start_time: {secs: 0, nsecs: 0}
duration: {secs: 10, nsecs: 0}"

Note: The simulated robot behaves differently from the real PR2. Maybe the F/T sensors change the kinematics and have to be added to the Gazebo model.

Collecting data

Starting Experiment A after modifying pr2_controller.yaml:

roslaunch ice_robot_controllers adaptNeuroController.launch
rosservice call /pr2_adaptNeuroController/runExperimentA "value: 0.5"

Saving data:

rostopic echo -p /pr2_adaptNeuroController/experimentDataA > datafile.rtp
rosservice call /pr2_adaptNeuroController/publishExpData "{}"

In future, the ice_experimenter will automatically save the data in a rosbag.

Topics

rostopic list | grep pr2_adapt
rostopic echo /pr2_adaptNeuroController/state/x -c
rostopic echo /pr2_adaptNeuroController/state/x_desi -c

sven-cremer/skinlearn-dev