The CoWriter project: Teaching a robot to write

A set of ROS nodes which facilitate the user interaction allowing a robot to be taught handwriting.

An example result of the interaction achievable with cowriter_letter_learning: Words are requested by showing cards to the robot with chilitags on them; children correct the robot's simulated handwriting by providing demonstrations on a tablet, which are used to update the robot's shape learning algorithm.

Dependencies

shape_learning for the letter learning algorithm,
nao_writing for the synchronised trajectory tracing capabilities for the robot,
ROS for Android for compiling the tablet app,
ros_markers for detecting the word cards,
gscam for using an external webcam to detect word cards instead of the Nao's camera.

For more information, see the readmes in internal directories and the parameters which are available for specification in the launch files.

Usage

With only a tablet

(With the shape_learner app deployed on the tablet)

export ROS_IP=(computer's IP)
roslaunch letter_learning_interaction nao_learning.launch use_robot_in_interaction:=false

rostopic pub /words_to_write std_msgs/String "use" -1 to send words to write (e.g. 'use') manually.

Note that console output can be viewed with rosrun rqt_console rqt_console.

An example of the system learning the word 'cow' (blue) based on user demonstrations with the tablet (green).

With a webots simulated Nao running

(With the shape_learner app deployed on the tablet, a webcam plugged in, and word cards with fiducial markers printed)

export ROS_IP=(computer's IP)
roslaunch letter_learning_interaction nao_learning.launch

(The camera device may be specified by appending camera_device:=/dev/video1, for example.) (Alternatively, rostopic pub /words_to_write std_msgs/String "use" -1 may be used to send words to write (e.g. 'use') manually, without detecting cards.) Note that console output can be viewed with rosrun rqt_console rqt_console.

With a ROS-enabled Nao

(With the shape_learner app deployed on the tablet and word cards with fiducial markers printed (robot's camera will be used))

Install chrony on the computer so that the robot may sync its clock.

On the robot:

sudo /etc/init.d/ntpd stop
sudo ntpdate (computer's IP)
sudo /etc/init.d/ntpd start

ntpdate -q (computer's IP) should then give ~0 offset, indicating that the clocks are synchronised.

export ROS_MASTER_URI=http://(computer's IP):11311
roslaunch nao_bringup nao.launch

On the computer acting as the ROS master:

export ROS_IP=(computer's IP)
roslaunch letter_learning_interaction nao_learning.launch use_sim_nao:=false nao_ip:=(nao's IP) use_external_camera:=false