Type: Bachelor or Master Thesis

 

Description: Most daily tasks involve interaction with the environment which imposes constraints on user's movement during execution. Drawing on a rough surface, for example, perturbs movement and appropriate forces need to be exerted to counteract disturbances and follow desired path. Drawing on a different material would require different skills although the task itself which is drawing of a shape, for example, remains the same. Each demonstration thus, consists of a component describing the task itself and a component desribing the contraints imposed by the environment. We are interested in learning a task by learning the motion and force policies for this task from variable-constraint demonstrations. The human demonstrates a task under variable contexts and the pure task policy is extracted by direct policy learning [1], [2].

 

Tasks:
• Design of the scenario {decide the manipulation task and capture data by human demonstration }
• Implementation of a direct policy learning algorithm [1] to learn a task’s movement and force
generation policy.
• Learning motion and force polices from variable-constraint demonstrations.
• Testing generalization of learned policies to different contexts.

Requirements:

Very good knowledge of C++ and Matlab.

Bibliography:
[1] Matthew Howard, Stefan Klanke, Michael Gienger, Christian Goerick and Sethu Vijayakumar.
A Novel Method for Learning Policies from Constrained Motion. In Proc. IEEE International
Conference on Robotics and Automation (ICRA), 2009.
[2] Chris Towell, Matthew Howard and Sethu Vijayakumar. Learning Nullspace Policies. In Proc.
IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS), 2010.

 

Supervisors: MSc Vasiliki Koropouli, Prof. Dongheui Lee.