Nicolas Durand

Automatic aircraft conflict resolution using genetic algorithms

In this paper, we show how genetic algorithms can be used to solve en-route aircraft conflict automatically to increase Air Traffic Control capacity in high density areas. The ATC _ background and the model are presented. The complexity of the problem is then discussed. The author then justifies the choice of GAs. After a brief description of genetic algorithms, the author describes the improvements that were used for solving the conflict resolution problem. Several numerical applications are then given justifying the choices that were made and illustrating the interest of the research. Next steps of this work are discussed in the conclusion.

Neural Nets Trained by Genetic Algorithms for Collision Avoidance

As air traffic keeps increasing, many research programs focus on collision avoidance techniques. For short or medium term avoidance, new headings have to be computed almost on the spot, and feed forward neural nets are susceptible to find solutions in a much shorter amount of time than classical avoidance algorithms (A*, stochastic optimization, etc.) In this article, we show that a neural network can be built with unsupervised learning to compute nearly optimal trajectories to solve two aircraft conflicts with the highest reliability, while computing headings in a few milliseconds.

Genetic algorithms applied to airport ground traffic optimization

Due to air traffic growth and especially hubs development, major European airports can easily become bottlenecks in the global air transportation network. Therefore, accurate models of airport traffic prediction become more and more necessary for ground controllers. In this paper, a ground traffic simulation tool is proposed and applied to Roissy Charles De Gaulle airport. Two global optimization methods, using genetic algorithms at the airport, are developed to minimize taxing time while respecting separation and runway capabilities. In order to compare the efficiency of the different methods, simulations are carried out on a one day traffic sample, and ground delay is correlated to the traffic density at the airport.

Turbo codes optimization using genetic algorithms

Turbo codes have been an important revolution in the digital communications world. Since their discovery, the coding community has been trying to understand, explain and improve turbo codes. The floor phenomenon is the parallel concatenated convolutional turbo codes main problem. In this paper, genetic algorithms are used to lower the free distance of such a code. Results in terms of bit error rate are compared to the other main methods.

CATS: A Complete Air Traffic Simulator

The main ideas behind the design of CATS were to have a general purpose simulator able to be a stable basis for the development of other modules, representing different levels of Air Traffic Control and Air Traffic Management. The basic framework and hypothesis of the CATS simulator are introduced. Two approaches of automatic conflict resolution are detailed as examples. The first one deals with a centralized control system for high density areas and medium term (10 to 15 minutes ahead) control. The second one deals with an on board reactive short term (3 to 10 minutes ahead) automatic conflict solver. As both solvers use a conflict detection that is robust to uncertainties, a convex modeling of uncertainties is first introduced.

Separating air traffic flows by allocating 3D-trajectories

This paper introduces two algorithms which allocate optimal separated 3D-trajectories to the main traffic flows. The first approach is a 1 vs. n strategy which applies an A* algorithm iteratively to each flow. The second is a global approach using a genetic algorithm, applied to a population of trajectory sets. The algorithms are first tried on a toy problem, and then applied to real traffic data, using operational aircraft performances. The cumulated costs of the trajectory deviations are used to compare the two algorithms.

Scheduling heuristics for on-board sequential air conflict solving

The resolution of conflicts between n aircraft is highly combinational and cannot be optimally solved using classical mathematical optimization techniques. Using a priority order to solve a n-aircraft conflict is much easier but the solution is not optimal. FACES (free-flight autonomous coordinated en route solver) is a model for coordinated on-board sequential conflict solving. In this project, conflict-free trajectories are obtained by applying elementary maneuvers to each aircraft, sequentially, according to some priority order. In this paper, a comparative study on a set of proposed priority heuristics to provide a suitable priority order is presented, and aggregated heuristics are compared according to some criteria. The conflict solver FACES using heuristics is tested in en-route upper airspace within an air traffic simulator using real traffic data.

OPTIMAL RESOLUTION OF EN ROUTE CONFLICTS.

In this article, we present mathematical modeling for en route conflict resolution, discuss the mathematical complexity of this problem, and show that classical mathematical optimization techniques are not suited to solve it. Instead, a stochastic optimization algorithm based on genetic techniques is presented. This algorithm can find many different nearly optimal solutions in real time, even in very complex situations involving many aircraft. Different examples of resolution with highly loaded traffic are presented and the results of the solver on an air traffic controller simulation prototype with real flight plans are evaluated. Limitations of the system and possible improvements are discussed.

A Genetic Algorithm to Improve an Othello Program

Let a computer program learn to play a game has been for long a subject of studies. It probably began in 1959 with A. Samuels program [Sam59], and similar methods are still used on the most advanced computer programs, such as Deep Thought ; however, in chess, the different methods used are mainly linear regression on parameters of the evaluation function. Othello was, from the start, a game where learning proved to be very useful. The BILL program used Bayesian learning [LM90] to improve parameters of the evaluation function. Similar methods are still applied for the best Othello programs (such as LOCISTELL0[Bur94]). However, these methods require large databases of games, and are very efficient only on Othello game playing, because the game always terminates in a fixed number of moves, with perfect end-games up to 15 moves. Such methods would be very difficult to use on chess playing algorithms. Other techniques used include co-evolution [SG94], use of neural networks trained by GA to focus minimax-search [MM94], evolving strategies with GA [MM93]. However, [SG94] did not lead to world class Othello program, [MM94] was unable to improve search as soon as the evaluation function was good enough to correctly predict the move, and [MM93] was apparently not able to improve the classical (positional+mobility) strategy. In this article, we show how genetic algorithms can be used to evolve the parameters of the evaluation function of an Othello program. We must stress that our method can be used on any algorithm using an evaluation function, for any two players game. In the first part of the article, we explain the structure of the Othello program. In the second part, we explain the structure of our genetic algorithm, the operators (crossover, mutation) used, and our method to compute fitness. In the third part, results are presented and, in the last part, possible improvements and generalization of this method are discussed

Finding and proving the optimum: cooperative stochastic and deterministic search

In this article, we introduce a global cooperative approach between an Interval Branch and Bound Algorithm and an Evolutionary Algorithm, that takes advantage of both methods to optimize a function for which an inclusion function can be expressed. The Branch and Bound algorithm deletes whole blocks of the search space whereas the Evolutionary Algorithm looks for the optimum in the remaining space and sends to the IBBA the best evaluation found in order to improve its Bound. The two algorithms run independently and update common information through shared memory. n nThe cooperative algorithm prevents premature and local convergence of the evolutionary algorithm, while speeding up the convergence of the branch and bound algorithm. Moreover, the result found is the proved global optimum. n nIn part 1, a short background is introduced. Part 2.1 describes the basic Interval Branch and Bound Algorithm and part 2.2 the Evolutionary Algorithm. Part 3 introduces the cooperative algorithm and part 4 gives the results of the algorithms on benchmark functions. The last part concludes and gives suggestions of avenues of further research.