Planning and Control of Multi-Agent Systems under Signal Temporal Logic Specifications
Time: Fri 2020-10-16 15.00
Location: zoom link for online defence. If you lack a computer or computer skills, please contact firstname.lastname@example.org for information. (English)
Subject area: Electrical Engineering
Doctoral student: Lars Lindemann , Reglerteknik, Division of Decision and Control Systems
Opponent: Professor Antoine Girard, Centre national de la recherche scientifique (CNRS)
Supervisor: Dimos V. Dimarogonas, Reglerteknik, Centrum för autonoma system, CAS, ACCESS Linnaeus Centre
Engineered systems are becoming more connected due to the availability of affordable and mobile communication and computation devices and form, already today, interacting and dependent complex networks. Examples of such systems range from multi-robot systems, e.g., fleets of drones or autonomous cars, over smart grids and smart home technologies. Besides these systems being connected and dependent, they have to satisfy complex individual and global specifications. These specifications may in particular include nested combinations of temporal and spatial requirements, e.g., a fleet of drones may be required to repeatedly inspect particular areas of interest within certain time intervals and change its formation pattern over time, while ensuring safety at all times. The challenge in such networks under complex specifications is to formally verify and ensure the correct and safe behavior of the system.
This thesis proposes planning and feedback control algorithms towards achieving this goal. In particular, we consider coupled multi-agent systems under signal temporal logic specifications. In the first part of the thesis, we propose timed automata-based planning algorithms that decompose the complex specification into a sequence of simpler specifications that can be realized by feedback control laws, such as those proposed in the second part of the thesis. These planning algorithms take uncontrollable events such as failures of agents/sensors into account and address perception issues, e.g., when the environment is not perfectly known. The second part proposes feedback control laws for multi-agent systems under fragments of individual and global signal temporal logic specifications. We propose time-varying feedback control laws that induce a desired temporal behavior, according to the specification at hand, into the closed-loop of the system. The combination of timed automata-based planning and feedback control is efficient in the sense that the computationally expensive planning algorithms can be performed offline, while only the efficient low-level feedback control laws are used during runtime of the system. The proposed algorithms additionally ensure robustness of the system. All theoretical results are illustrated in simulations or experiments involving mobile robots.