Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Robust adaptive model predictive control of quadrotors
Designing robust controllers for real-life systems results in conservative control performance. This conservatism can be reduced by learning the model online and tuning the controller. In this project, a robust adaptive MPC scheme will be used to deliver packages reliably with a quadrotor.
Keywords: model predictive control, adaptive control, parameter estimation, safe learning
Model predictive control (MPC) techniques are popular for their ability to guarantee stability and constraint satisfaction. However, since the stability of the controller depends explicitly on the model accuracy, robust MPC techniques are used to handle modeling uncertainties. When the uncertainty involved is due to constant or slowly varying parameters, the performance of robust MPC methods is conservative. This conservatism can be reduced by using adaptive strategies that can learn the unknown parameters online. Robust adaptive MPC is one such strategy, which has received a lot of attention in the recent literature.
This project aims to apply this technique to a quadrotor with parametric uncertainty in its model. Such uncertainty would be common, for example in the case of a drone delivery service where only the maximum payload is known apriori. The effect of wind can also be modeled as a time-varying uncertainty. A parametric model of the drone would be developed as a part of the thesis, with multiple uncertain parameters. Then, a robust adaptive MPC controller will be developed, and its performance will be compared to other methods in the literature through simulations and experiments.
Model predictive control (MPC) techniques are popular for their ability to guarantee stability and constraint satisfaction. However, since the stability of the controller depends explicitly on the model accuracy, robust MPC techniques are used to handle modeling uncertainties. When the uncertainty involved is due to constant or slowly varying parameters, the performance of robust MPC methods is conservative. This conservatism can be reduced by using adaptive strategies that can learn the unknown parameters online. Robust adaptive MPC is one such strategy, which has received a lot of attention in the recent literature.
This project aims to apply this technique to a quadrotor with parametric uncertainty in its model. Such uncertainty would be common, for example in the case of a drone delivery service where only the maximum payload is known apriori. The effect of wind can also be modeled as a time-varying uncertainty. A parametric model of the drone would be developed as a part of the thesis, with multiple uncertain parameters. Then, a robust adaptive MPC controller will be developed, and its performance will be compared to other methods in the literature through simulations and experiments.
1) Review literature related to existing methods for adaptive control, online parameter estimation and learning-based MPC.
2) Develop a quadrotor model with parametric uncertainty for different scenarios of a drone-based delivery system
3) Implement robust adaptive MPC to control a quadrotor, and compare its performance to other methods using simulations and experiments.
1) Review literature related to existing methods for adaptive control, online parameter estimation and learning-based MPC.
2) Develop a quadrotor model with parametric uncertainty for different scenarios of a drone-based delivery system
3) Implement robust adaptive MPC to control a quadrotor, and compare its performance to other methods using simulations and experiments.
Anil Parsi, aparsi@control.ee.ethz.ch
Jeremy Coulson, jcoulson@control.ee.ethz.ch
Prof. Roy Smith, rsmith@control.ee.ethz.ch