Back to the main AGENTFLY page

The project is aimed at large-scale distributed fast-time simulation and control of civilian air traffic over the National Air Space (NAS) of the United States. It also serves as a tool for validation of algorithms proposed for the NextGen concept.

The implementation relies on the underlying simulation system. Its architecture has been designed and developed specifically to support execution of large-scale simulations featuring tens of thousands of airplanes. The system employs methods of automatic load balancing in realtime which ensure an efficient execution of the distributed simulation on the given number of computers. NextGen AGENTFLY has been successfully used for simulation of more than 50,000 airplanes flying within the National Air Space, based on real air traffic data spanning one day provided by FAA. During peek hours of the simulated air traffic, there were around 6,000 airplanes being managed at once.

Airplanes modeling supports simulation over the entire airspace of the Earth with positions of airplanes described by their GPS coordinates. The system has been integrated with real-world data describing current wind, as well as geometry and time constraints of Special Use Airspaces (SUAs) which the airplanes are not allowed to enter during their flight. The airplane performance model is based on Base of Aircraft Data (BADA) specifications maintained by EUROCONTROL.

Each simulated airplane can use one of two different behaviors. Airplane can behave as standard IFR flight following instructions of air traffic controller or it can be switched to autonomous mode where is an agent responsible for flight path planning for the particular airplane, and negotiation with other airplanes (i.e. the corresponding agents) over detected conflicts in planned trajectories, which allows safe, collision-free operation of multiple airplanes within the same air space. In order to ensure safety of all flights and prevent separation violations, conflicts of airplanes’ flight paths are resolved automatically either in a cooperative or non-cooperative manner, depending on the ability and/or willingness of the involved airplanes to communicate.

The AGENTFLY system supports modeling of air traffic controller behavior. The ATC model is based on VCAP (visual, cognitive, auditory, psychomotor) model, where each action is assigned to use some of VCAP resources. The behavior is based on procedures (e.g. handoffs, collsion detection and resolution, clearance instructions, etc.). Each procedure is decomposed into atomic actions (e.g. talking to pilot using radio, listening to pilot, typing into keyboard, thinking, etc.). Each action has assigned (deterministically randomized) duration, used VCAP resources and workload for each resource. Procedures as triggered by the simplified implementation of ERAM system. Duration of procedures can cause delays in execution and it can lead to overload of the air traffic controller. The ATC model has been validated against human-in-the-loop experiment performed by FAA.

The system also provides built-in tools for realtime collection and subsequent evaluation and analysis of behavior of the system under various conditions. This has enabled us to perform measurements and experiments analyzing a number of air-traffic situations and system configurations.

The project has been supported by the U.S. Federal Aviation Administration (FAA), U.S. Air Force Research Laboratory (AFRL) and by the Czech Ministry of Education, Youth and Sports.

Back to the main AGENTFLY page

Project Videos

(0:57 12MB)
(1:05 50MB)
(1:08 319MB)
(1:09 256MB)
(0:51 29MB)
(0:48 5MB)



  • David Sislak and Premysl Volf and Michal Pechoucek and Christopher T. Cannon and Duc N. Nguyen and William C. Regli: Multi-Agent Simulation of En-Route Human Air-Traffic Controller. In Proceedings of the Twenty-Fourth Innovative Appications of Artificial Intelligence Conference. Toronto, Canada: AAAI Press, 2012, p. 2323-2328. ISBN 978-1-57735-568-7.
    BiBTeX | PDF (382)
  • David Sislak and Premysl Volf and Dusan Pavlicek and Michal Pechoucek: AGENTFLY: Multi-Agent Simulation of Air-Traffic Management. In 20th European Conference on Artificial Intelligence. Montpellier, France: IOS Press, 2012. ISBN 978-1-61499-097-0.
    BiBTeX (383)


  • David Sislak, Premysl Volf, Stepan Kopriva and Michal Pechoucek: AgentFly: Scalable, High-Fidelity Framework for Simulation, Planning and Collision Avoidance of Multiple UAVs. In Sense and Avoid in UAS: Research and Applications. Wiley: John Wiley&Sons, Inc., 2012, p. 235-264.
    BiBTeX (315)