Implementation of the Cerebellar Model Articular Controller neural network and its application on the robot NAO for following a ball with the robots arm. The training and the testing is made at a fixed robots head position and allowing only two joints from the arm to move (elbow and shoulder).

The training data is collected by putting the ball in NAO's hand and saving both ball position in the camera's image and shoulder and elbow joints angles.

After the training is done, the robot will follow the ball continously as long as it appears in its field of view.

A video demo can be found under this link:
https://www.youtube.com/watch?v=Wfy7c-djeAs


Files Included:
central_node.cpp: A node handling all the joind and image data coming from the robot. We were given a squeleton as introduction to ROS and changed it, adding all the needed functions.

cmac.h: Implementation of the Cerebellar Model Articular Controller

training_data.txt: collected training data that can be used directly.


These are only the source codes of a node contained in a package "BILHR_ros" having all needed messages for NAO installed (Check tutorial on ros.org about how to install ROS drivers for NAO).