Skip to main content

Coordination of Multi-Agent Systems: Predictive and Vision-based Control for Aerial and Space Robotics

Time: Fri 2022-03-18 17.00

Location: Q2, Malvinas väg 10, Stockholm

Video link:

Language: English

Doctoral student: Pedro Roque , Reglerteknik, Royal Institute of Technology

Opponent: Prof. Soon-Jo Chung, California Institute of Technology (Caltech)

Supervisor: Dimos V. Dimarogonas, Reglerteknik; Mikael Johansson, Reglerteknik

QC 20220224


Aerial and space autonomous systems are safety-critical systems that need to operate safely and reliably for long periods of time. Their applications range from autonomous inspection of complex and sensitive structures, to surveillance and transportation of sensitive payloads. These tasks impose strict operation requirements in order to be successfully accomplished, which the operating agents must ensure at all times while taking advantage of all the available sensing and communication modules, but considering energy constraints and individual autonomy as much as possible. 

In this thesis, predictive and vision-based control techniques are explored to coordinate multi-agent systems, with a particular focus to aerial and space robotic applications. A particular focus is set on model predictive control (MPC) and image-based visual servoing (IBVS), and extensions are provided for controlling multi-agent systems using these frameworks. Initially, MPC is explored to distributively navigate a formation of multiple unmanned aerial vehicles (UAVs) in a leader-follower scheme, where minimal communication among agents in the formation is prioritized. Then, MPC is explored with IBVS in a cascaded control fashion to stabilize a UAV using image features, while taking advantage of the optimal control inputs provided by the MPC module given the dynamics of an aerial vehicle. This work serves as motivation and stepping stone to two other extensions. The first is a distributed IBVS framework to coordinate multiple agents using image-features and one distance measurement between any 2 agents in the formation, removing the need for magnetometers or global reference frames. The second is a robust MPC controller for tracking of time-varying trajectories considering additive and bounded noise on the system dynamics and safety of the agent. The thesis concludes with the main ideas provided in the manuscript, and sheds light on the possible future research directions that may spawn from this work.