Skip to main content

Model Predictive Control for Cooperative Multi-UAV Systems

Examiner Dimos Dimarogonas

Time: Mon 2021-06-14 11.00 - 12.00

Location: Zoom-link:

Doctoral student: Roberto Castro Sundin , DCS - Reglerteknik

Opponent: Johanna Andersson

Supervisor: Pedro Roque

Abstract: The maneuverability and freedom provided by unmanned aerial vehicles (UAVs) make these an interesting choice for transporting objects in settings such as search and rescue operations, construction, and smart factories. A commonly proposed method of transport is by using cables attached between each UAV and the payload. However, the geometrical constraints posed by these attachments typically result in a system with highly complex dynamics. Although not an issue for conventional PID control schemes, these complex dynamics make the direct application of model predictive controllers (MPCs) infeasible for real-time usage. For this reason, much of the previous work has focused on treating the payload as a disturbance, thereby losing the ability to predict its effect on the UAVs. Contrary to this, this thesis presents an MPC that both captures the dynamics of the payload, and is capable of real-time usage. This is made possible by a parametrized linearization of the original system, and results in greatly improved performance compared to the disturbance model approach. The controller is derived for a system with two UAVs that transport a bar-like payload and verified both in simulations and physical experiments. The resulting control system is able track a multitude of setpoints, including rotations of both payload and UAVs, as well as lateral translations. Furthermore, it is able to attenuate external disturbances well, and dampens and prevents oscillations more efficiently when compared to the disturbance based approach. The resulting MPC solving time is on the order of milliseconds. Additionally, an initial attempt to decentralize the system is made, and the resulting controller experimentally tested on the UAV--bar system, resulting in a lower MPC solving time (2.5 times faster on average), but worsened performance in terms of position tracking of the bar.

Page responsible:Web editors at EECS
Belongs to: Decision and Control Systems
Last changed: Jun 09, 2021