Learning from Demonstration

In the industry, there has been a constant demand for methods that allow for a time- and cost-efficient re-programming of robots by non-expert personnel. Research on task-level programming (TLP), robot skills and Learning from Demonstration (LfD) has produced promising results but many approaches still rely on large amounts of data or require the end-users to have experience in the handling of robots. We aim to build a framework, which achieves the extraction of a parametrized skill sequence from a video recording of an assembly task demonstration executed by a human demonstrator.

Open Topics

• A thesis can be defined individually. Please send me an email.

Robotic Skills and Grasping Skills

Here we aim to use the extracted skill sequence from above LfD framework to execute the task on a new environment. Therefore, we need to implement intelligent robot skills. In addition, physical parameters of the demonstration must be learned, as these are not determined by pure passive observation. We give the robot a sense of touch through additional sensors in the gripper and develop algorithms to identify physical parameters of the environment.

Open Topics

• A thesis can be defined individually. Please send me an email.