Robot Learning by Demonstration with Local Gaussian Process Regression. In Proc. 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems, Taipeh, Taiwan, 2010. Learning from Demonstration using a Katana robotic manipulator. Note that in each demonstration the objects' positions change. Learning means to generalize from these training samples to an arbitrary new situation where all the objects can be located at different positions. In the reproduction, the manipulator's trajectory is computed based on the constraints extracted from the recorded demonstrations. Unlike in classical teach-in approaches, this method is able to deal with changing objects' positions.