Sprungmarken

Servicenavigation

Hauptnavigation

Sie sind hier:

Hauptinhalt

Detailseite



Online Motion Planning and Control of Cooperative Robotic Manipulators

Lehrstuhl: Institute of Control Theory and Systems Engineering

Betreuer: Freia Irina John, Frank Hoffmann,

Beginn ab: 20.04.2017

Maximale Anzahl der Teilnehmer: 7

Beschreibung: Cooperative robots help increasing productivity and assist workers. A collaborative robot physically interacts with humans in a shared workspace. The project group is concerned with kinesthetic teaching of compliant robots, encoding of skills and reproduction of the motion under kinematic and dynamic constraints as well as motion planning and control in joint and task space.

Learning from Demonstration is an established paradigm to teach robot skills in an intuitive manner. A statistical model encodes the demonstrated movements performed by the human. This model is utilized for reproduction under different situations or with diverse architectures. The motion task is encoded with a Gaussian Mixture Model (GMM) which captures the probability of a data sample belonging to the underlying distribution of demonstrations.

The reproduction of the intended skill rests upon Gaussian Mixture Regression (GMR), which not only provides a reference trajectory but also captures possible variations of the task. However, the robot might not be able to exactly reproduce the designated motion due to (kino-)dynamic constraints, obstacles or disturbances. This observation demands a control scheme that reconciles the execution of the nominal task with the kinematic and dynamic capabilities of the robot and constraints imposed by the workspace.

The project group addresses the following problem: Reproduce a motion demonstrated by a teacher in the most faithful manner by a robot which limited kino-dynamic capabilities prohibit a perfect imitation of the skill.

Model Predictive Control (MPC) is efficient in generating an optimal trajectory with respect to an underlying objective function subject to different constraints. The combination of MPC and GMR provides a mean to attain the faithful reproduction of the skill with constraints and additional objectives of the robots motion capabilities.

Participants are supposed to design and implement schemes for learning from demonstration, task reproduction and online motion planning in ROS (Robot Operating System) for the cooperative robots UR10 and Sawyer at the IRF. The schemes are evaluated in experiments on the real robots.


Students are expected to have a background in robotics, control theory and optimization. They are also expected to have either completed or at attend the two modules on robotic manipulators and mobile robots with ROS:
- Mobile robots
- Modelling and control of robotic manipulators.
Students are also expected to have a profound programming experience in C++ and/or Matlab. Knowledge about ROS is a clear advantage. Registration starts on 14.01.2017.

* C. Rösmann, F. Hoffmann, and T. Bertram, “Timed-elastic-bands for time-optimal point-to-point nonlinear model predictive control”, Control Conference (ECC), 2015 European, pp. 3352–3357, July 2015.
** S. Calinon, “Robot programming by demonstration: a probabilistic approach”, EPFL Press ISBN 978-2-940222-31-5, CRC Press ISBN 978-1-4398-0867-2, 222 pages, hardcover, 2009.
*** Myrel Alsayegh, Christoph Rösmann, Frank Hoffmann, Model Predictive Control for Learning from Demonstration, Computational Intelligence Workshop, December 2016.
**** www.ros.org