Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Data-Enabled Predictive Control of Robotic Systems
As systems are becoming more complex and data is becoming more readily available, scientists and practitioners are beginning to bypass classical model-based techniques in favour of data-driven methods. We wish to explore the world of data-driven methods through implementing a data-driven control algorithm on a real robotic system.
Keywords: control; robotics; data-driven
The focus of this project will be to implement a data-driven algorithm in order to control a simple robotic system without assuming knowledge of the system model. The system on which the algorithm will be tested consists of a single actuator with a pole with modifiable inertia. In particular, we will ask the pole to optimally track a desired trajectory while respecting desired constraints. Without a system model, we aim to synthesize optimal control inputs in real-time only using measured data from previous trajectories. More information can be found here: https://arxiv.org/abs/1811.05890
The focus of this project will be to implement a data-driven algorithm in order to control a simple robotic system without assuming knowledge of the system model. The system on which the algorithm will be tested consists of a single actuator with a pole with modifiable inertia. In particular, we will ask the pole to optimally track a desired trajectory while respecting desired constraints. Without a system model, we aim to synthesize optimal control inputs in real-time only using measured data from previous trajectories. More information can be found here: https://arxiv.org/abs/1811.05890
The student will become familiar with both model-based and data-driven control methods by simulating them on the aforementioned system. The student will then implement a data-enabled predictive control algorithm on the real system. This algorithm can serve as a starting point for further theoretical developments, and comparisons between model-based and data-driven control methods.
Work packages:
- Implement algorithm (currently MATLAB) in C++ to be able to use it with the real system
- Test implementation in simulation and on the hardware
- Compare the algorithm with model-based methods
The student will become familiar with both model-based and data-driven control methods by simulating them on the aforementioned system. The student will then implement a data-enabled predictive control algorithm on the real system. This algorithm can serve as a starting point for further theoretical developments, and comparisons between model-based and data-driven control methods.
Work packages:
- Implement algorithm (currently MATLAB) in C++ to be able to use it with the real system
- Test implementation in simulation and on the hardware
- Compare the algorithm with model-based methods
Pascal Egli, pasegli@ethz.ch
Jeremy Coulson, jcoulson@control.ee.ethz.ch