Anomaly Detection and Behavior Modelling

The anomaly detection aims to identify samples, which in some form deviates from the majority. I It is an optional part of the data-processing, which can improve the standard algorithms, or it is used at its own. An examples of applications are identification of measurement errors (faulty sensors), detection of faulty behavior of some industry process (anomaly might indicate a malfunction), fraud detection in credit card transactions, monitoring of health or environmental processes, etc. Most approaches to the anomaly detection are very computationally expensive, which makes their practical applicability questionable, especially in tasks requiring quick responses and constant update. Our research targets to make both parts (detection and update) real-time.Behavior modeling is a specific application of anomaly detection, where we search for behavior of individuals significantly different from the crowd. The applications are typically in security, e.g. terrorist attack prevention.

Contact person: Tomas Pevny

Learn more about Anomaly Detection and Behavior Modelling

Computational Game Theory

Game Theory is a formal framework for analyzing competitive situations to determine the optimal course of action for a self-interested agent. Computational game theory uses computational methods for modeling and solving game-theoretic problems. In our basic research, we focus on creating novel faster algorithms for solving games in standard representations, as well as creating new game-theoretic models that allow representing and solving specific classes of games more efficiently. In the applied research, we model specific real-world problems in the game-theoretic framework; we propose domain-specific improvements of the existing algorithms for computing (an approximation of) optimal solutions; and we evaluate the computed strategies in computer simulations. The problem domain we focused on include network security, transportation networks, and military operations.

Contact person: Viliam Lisy

Learn more about Computational Game Theory

Multi-agent Planning and Resource Allocation

In environments, where agents have their own goals, but they want to cooperate as well, it is necessary to look for common recipes, i.e., multi-agent plans. Using such plans, the agents fulfill the goals in a coordinated manner. Research field of multi-agent planning search for both theoretical guarantees and practical algorithms for solving of these problems.  Multi-agent planning problems can also comprise additional distinctive resources, which have to be handled by the agents. Resource allocation problems cover this extensions. The application domains can vary. In our group, we focus on a wide spectrum of problems from domain-independent planning, where the particular domain is a part of the input, to low-level domain specific planners, e.g., generating non-collision trajectories for hardware airplanes.

Contact person: Antonin Komenda

Learn more about Multi-agent planning and resource allocation in ATG

Large-scale Agent-based Modelling and Simulations

Agent-based modelling has become an indispensable tool for analysing the behavior of complex socio-technical systems. We perform basic and applied research as well as technology development in agent-based modeling and simulation. Our interests include developing flexible and scalable simulation toolkits, combining agent-based and discrete-event simulations, simulation development methodologies, and simulation-based experimental methodologies.  Our models typically have at least one of the following challenging properties: (1) complex, structured physical environments, (2) sophisticated cognitive agents and (3) very high number of agents (millions). Although much of our technology and know-how is application-agnostic, we have a strong application track in modelling air traffic management, tactical urban operations, maritime traffic and multi-modal transport systems.

 Contact person: Michal Jakob

Learn more about Large-scalae agent-based modelling and simulations

Autonomous Aerial Vehicles

Design and implementation of algorithms for coordination and cooperation in teams of autonomous robotic entities is no longer a matter of theoretical research only as their deployment in field scenarios grows rapidly. We thus not only design novel algorithms and approaches but also verify their features by deploying them on real hardware assets. At present we focus mainly on the area of unmanned aerial vehicles (UAVs), where we deploy algorithms on both fixed-wing assets and VTOLs. This allows us to verify validity of basic assumptions of designed algorithms in terms of communication and computational requirements, physical restrictions, uncertainty and robustness. Besides verification of distributed mission control algorithms we also develop new ones for 3D trajectory planning and collision avoidance. Within this research area we also focus on design of human-machine interfaces and integration of real aerial assets into mixed simulations together with simulated entities.

 Contact person: Milan Rollo

Agile Tactical Operations

Nowadays, the state of the art in robotics reached a point when troops in military operations theatre routinely deploy unmanned robotic assets, be it unmanned aerial and ground vehicles, or unattended ground sensors supporting them in carrying out their tactical missions. Tele-operation of large numbers of such assets drastically raises costs of such operations in terms of costly and scarce expert human resources.

Agile Tactical Operations aim at development and evaluation of multi-agent coordination techniques supporting information-collection missions in tactical urban warfare. The core objective of the cluster is development and evaluation of techniques allowing deployment of teams of autonomous assets which are capable of autonomous coordination and teamwork with the goal to carry out the high-level mission assigned to the team. In particular, we are developing agent-based coordination, planning and game-theoretic techniques for heterogeneous multi-agent teams allowing them to carry out various information collection tasks, such as reconnaissance, convoy and perimeter protection, pursuit-evasion of smart targets, surveillance, safe area traversal, etc.

Contact person: Jiri Vokrinek

Learn more about Agile Tactical Operations

Air Traffic Management

Air traffic is growing every year and it causes strong need to improve air traffic management. New systems and tools are developed to help air traffic controllers keep safety of airplanes. The validation of tools is crucial for their deployment. We are building framework to model and simulate air space environment, aircrafts performance models and cognitive models of air traffic controllers and pilots. The framework can be used for validation of new tools and concept in a world-wide scale.

Contact person: David Sislak

Critical Infrastructures Protection

We apply basic research in game theory, planning and adversarial reasoning to create abstract mathematical models of critical infrastructures (such as traffic systems, utility networks or communication infrastructures including computer networks) typically represented by a graph structure or a set of spatially distributed targets. The models are then used to design optimal protection policies with limited defensive resources, which is typically computationally expensive and requires research of novel and scalable algorithms. As the level of abstraction can be often substantial, we employ large-scale multi-agent simulations to assess robustness of the proposed solution in rich environments which capture additional properties of the physical world.

Contact person: Ondrej Vanek

Learn more about Critical Infrastructures Protection

Intelligent Transport Systems

Agent-based computing will play a fundamental role in intelligent, autonomous transport systems of the future. Such systems will capable of continuously and autonomously optimizing their operation so as to fulfil mobility needs of their users in most convenient, economical and environmentally-friendly ways. We study and develop a range of techniques that will underpin future transport systems, including agent-based modelling of multi-modal transport systems, personalized resource-aware journey planning, auctions and negotiation mechanisms for flexible transport markets, vehicle-to-vehicle coordination and cooperation techniques as well as game-theoretic optimization of transport security measures.

Contact person:  Michal Jakob

Learn more about  Agent-based Computing for  Intelligent Transport systems

Manufacturing and Logistics

The industrial cooperation is one of the most important application areas of Agent Technology Center. The goal of this activity is to apply state of the art data engineering practices and algorithmization know-how to real world problems encountered within partner enterprises, including production planning, production scheduling and logistics domain assistive tools. This effort fits well into what is currently perceived as the most feasible way of integrating advanced AI algorithms to industry expert driven applications - not by conceptually replacing human factor, but instead helping human operators in sub-tasks where a computer is massively more efficient than human brain. Examples of successful technology transfer are project-oriented capacity planning system of Modelarna Liaz pattern shop, engine batch production planning tool for Gedas CR, and advanced planning system for FOXCONN CZ supporting supply demand matching optimization and flexible production planning.

Contact person: Jiri Vokrinek

Learn more about Manufacturing and Logistics


Our contribution in the field of network security concentrates on the research of novel anomaly detection methods. In particular, we investigate the mechanisms for the detection of anomalous behavior in computer networks based on the processing of purely statistical information, without relying on the inspection of the content of the transmitted data. Due to the specific nature of the problem, we rely on ensemble approach, where the system consists of many unsupervised statistical anomaly detection algorithms, integrated by means of trust modeling, use to infer more stable long-term conclusions from the anomaly detector’s outputs. The system further relies on semi-supervised optimization framework based on the game theoretical optimization methods.

 Contact person: Martin Rehak


The steganograhy is the art of transmitting secret messages without raising any suspicion of secret communication taking place. This is usually achieved by hiding the message into innocuous looking objects, e.g. digital images. Steganalysis, is the opposite meaning that we want to detect the presence of hidden messages. Our goal is the development of approaches that works well in practice, which is very difficult due huge variety of images typically used by users. We have developed an unique approach estimating this variety from the observed images. Steganography is a joint project with Oxford University.

Contact person: Tomas Pevny