Sie sind hier:



Model Predictive Control for Robotic Pole Balancing

Lehrstuhl: Institute of Control Theory and Systems Engineering

Betreuer: Christoph Rösmann, Artemi Makarow, Maximilian Krämer,

Beginn ab: 12.10.2017

Maximale Anzahl der Teilnehmer: 8

Beschreibung: Pole balancing is a difficult task even for a human being,
especially if the pole is short, the available operation space is limited
or if some external disturbances occur. Hereby, the human profits from its profound
eye-hand coordination in terms of sensing and control.
A more sophisticated task is not to just stabilize the pole at the desired position but to also follow a predefined trajectory, e.g. a circular or rectangular.

Now imagine, you should teach a robot to perform this task autonomously.
Usually, a robotic arm is composed of a sequence of joints respectively servo drives
which generate the actual motion of the arm. For serious manipulation tasks, external sensing
e.g. cameras are required to obtain information about the environment.
Only with a suited mathematical framework at the control stage advanced tasks like pole balancing under constraints (limited space, maximum speed/acceleration) might be accomplished.

A promising modern control scheme to handle this task is Model Predictive Control (MPC).
MPC uses a (possibly simplified) model of the system dynamics to predict the future state evolution and determines an optimal control policy subject to a user-defined objective like minimum control effort or time optimality. Furthermore, MPC handles the previously mentioned constraints on states and control variables. In general, MPC is currently the most or one of the hottest topics in research and conquers more and more industrial applications. Common applications include process control, automated driving, and robotics.

The objective of the project group is to realize the MPC based pole balancing task in the following manner:
- Realization of the technical setup (camera based object tracking, robot communication)
- Development and implementation of conventional controllers which are well known in the industry (PID, LQR, ...)
- Development and implementation of a simple MPC concept cascaded with a robot motion controller
- Development and implementation of a holistic MPC including both the pole and the arm dynamics

All subtasks should be realized both in simulation and with a real industrial robot.
Optionally, or depending on the number of interested students, a second experiment in addition to pole balancing might be developed and realized (e.g. ball-on-a-plate).
Note, it is not required that participants are familiar with concepts in robotics or deep understanding of control theory. But they should be highly interested in control theory, mathematics, and programming.

Benefits for participants:
- Learning, understanding, and implementation of cutting-edge control concepts (which are important for a various range of future automated applications)
- Investigating (trajectory) optimization strategies
- Working with a real experimental system, especially a robotic arm
- Systematic approaches for scientific problem formulations
- A major improvement in modern C++ and/or Python/Matlab skills
- Robot Operating System (ROS) skills (for the robot's base controller and simulation)
- Introduction to modern software development (continuous integration, git)