Many next-generation air traffic algorithms may be formed by learning algorithms or dynamic programming techniques. These techniques form their solutions through iterative methods where the efficacy of a proposed solution needs to be evaluated for every round of iteration. In complex air traffic scenarios, often the only way to evaluate a proposed solution is to simulate the impact of the solution in an air traffic simulator. Such a simulator has to be fast enough to allow for many rounds of iteration. In addition, the simulator will have to be modular enough to allow modules to be created that simulate a portion of the airspace in detail.
Learning algorithms and dynamic programming methods may require thousands of rounds of iteration before converging on a satisfactory solution. If each solution is evaluated using a simulation, the complete simulation will have to be performed thousands of times. Therefore, each simulation will have to be performed quickly. While low-fidelity simulations can be performed quickly, high-fidelity simulations will be much slower. Depending on the problem domain, a learning algorithm will likely focus on certain aspects of the airspace. These parts of the airspace need to be simulated at high fidelity. At the same time, the rest of the airspace still needs to be simulated at low fidelity so that the aggregate impact of the solution proposed by the learning algorithm can be evaluated.
MFS (Multi-Fidelity Simulator) is a pluggable framework for creating an air traffic flow simulator at multiple levels of fidelity. The framework is designed to allow low-fidelity simulations of the entire U.S. airspace to be completed very quickly (on the order of seconds). The framework allows higher-fidelity plug-ins to be added to allow higher-fidelity simulations to occur in certain regions of the airspace concurrently with the low-fidelity simulation of the full airspace.
This work was done by Guilaume Brat of Ames Research Center, Adrian Agogino of the Regents of the University of California Santa Cruz, and Ritchie Lee of SGT Inc. This software is available for use. To request a copy, please visit here .