LARGE-SCALE HIGH-FIDELITY AGENT-BASED SIMULATION IN AIR TRAFFIC DOMAIN

Abstract

We present the concept of dynamic partitioning of scalable, high-fidelity multi-agent simulation complemented with intelligent load-balancing processes. The simulation framework is designed to simulate entities to high details that require extended computation resources. To be able to simulate a huge amount of entities, distributed simulation is introduced using spatial partitioning and dynamic load balancing. A novel and important feature is the combination of the synchronous and asynchronous parts in the simulation. We use the domain of the air traffic simulation to verify the simulation framework. We present a method to perform spatial and temporal planning within 3D space and multilayer architecture using several collision avoidance algorithms to illustrate the high computational demands of each airplane. The platform has been used to support simulation of an entire civilian air traffic touching the national airspace of United States. A thorough evaluation of the system has been performed, confirming that it can scale up to a very high number of complex agents operating simultaneously (thousands of aircrafts) with full detailed models.

KEYWORDS:

• agent-based simulation
• air traffic
• collision avoidance
• large-scale simulation
• trajectory planning

Acknowledgments

The work has been supported by the Federal Aviation Administration (FAA) under project number DTFACT-08-C-00033 and by the Czech Ministry of Education under grant number 6840770038. The underlying AGENTFLY system was supported by the Air Force Office of Scientific Research, Air Force Material Command, USAF, under grant number FA8655-06-1-3073. The views and conclusions contained herein are those of the author and should not be interpreted as representing the official policies or endorsements, either expressed or implied, of the Federal Aviation Administration, the Air Force Office of Scientific Research, or the U.S. Government.

Notes

¹Air traffic touching national airspace of the United States consists of all airplanes flying at least part of their trajectories inside national airspace.

²The proposed modular architecture is domain independent. Therefore, it is ready for deployment on autonomous vehicles like airplanes or ground vehicles (VFRs; Visual flight rules). The implementation in AgentFly is used as a reference mechanism to test the efficiency of the following three CA algorithms. The iterative peer-to-peer collision avoidance (IPPCA) algorithm employs pair-wise negotiation to solve the collision. The multiparty collision avoidance (MPCA) algorithm extends the first two algorithms by allowing several assets to negotiate over their collective CA maneuvers. The noncooperative collision avoidance (NCCA) algorithm supports CA when communication between assets is impossible. This class of algorithms is based on modeling and predicting the noncooperative object's future airspace occupancy and representing its possible future positions in terms of dynamic no-flight zones.

³By situated entities we mean that they are embedded in a synthetic model of a 3D physical space.