Escolar Documentos
Profissional Documentos
Cultura Documentos
54
London, UK, May 29-31, 2012 ATACCS’2012 | RESEARCH PAPERS
In this paper, an air traffic conflict resolution problem (see Figure 2). Therefore the algorithm based on the simple
under heavy traffic conditions is studied. It is assumed that discretization of airspace in cells does not ensure minimum
the initial trajectories of each aircraft are known and the separation between aircraft. A natural idea to cope with this
detection method is based on a grid model [15]. Therefore, disadvantage is simply to change the conflict definition
each trajectory is defined by the cells through which the [20]. Therefore, the following definitions are considered in
aircraft passes. The proposed method is based on speed this paper:
planning to solve conflicts. The speed profiles for all the Neighboring cells to C(i): is the set of cells whose
aircraft involved in a conflict are computed and the separation from C(i) is less than the minimum separation.
problem is solved by maintaining the planned 4D
trajectory. This approach has the advantage that the Conflict: C(i) is crossed by an aircraft is in conflict if
probability of creating new conflicts with other aircrafts in there is another aircraft that crosses a cell in the
the airspace is low. neighborhood of C(i).
The conflicts between two or more aircraft are solved in Conflict Zone: Let C1(n) and C2(n) be the sequence of
real time, at low computational cost, once they are detected. cells visited by AV1 and AV2 (see Figure 1). Let us
A related paper is [19] in which a collision avoidance assume that C1(C5) and C2(E3) are the first conflictive
method for multiple aerial vehicles (AVs) sharing airspace. cells between them. Eventually C1(G5) and C2(E7) are
The proposed method is based on the technique presented also in conflict while C1(H5) and C2(E8) are not. Then,
in [19], in which suboptimal solutions are found. The the conflict zone is composed of the cells
proposed method ensures separation between aircraft, C1(C5;D5;E5;F5;G5;H5) and C2(E3;E4;E5;E6;E7) (see
which is not the case for the method described in [19]. Figure 1). This definition could be easily extended to
conflict zones crossed by more than two aircraft.
This paper is organized into seven sections. The
formulation of the problem is described in Section II. The Collision: two aircraft cross a conflict zone at the same
description of the proposed CDR method is presented in time.
Section III. Finally, the simulations and experiments
carried out are covered in Section IV and the conclusions
are detailed in Section V.
PROBLEM FORMULATION
The problem considered in this paper concerns collision
detection and resolution among multiple aircraft under
heavy traffic conditions. In order to maintain the safety
distance and previously computed planned trajectories, the
heading changes are not allowed and the solution only
considers speed changes. Each aircraft computes its
collision-free trajectory by changing cooperatively its speed
profile. This speed profile presents an improvement with
respect to [9] where only an aircraft performs the maneuver
to avoid the conflict.
The detection algorithm is based on the discretization of the
airspace divided into cubic cells, also called the grid model
(see Figure 1). The discretization of the airspace offers an
advantage: the detection algorithm is simpler and faster
because a trajectory is parameterized by the number of cells Figure 1: Scenario with two aerial vehicles and conflict
that the aircraft passes through with entrance and departure zone (gray cells). The safety distance is given by two cells.
time. The safety distance or minimum separation is given We distinguish three types of aircraft in order to solve
by a number of cells and the size of cell is set between five situations where a solution that avoids a collision generates
or ten times lower than the safety distance in order to a new collision with other aircraft. These types are: aircraft
minimize the discretization effects while not increasing the that is directly involved in the detected collision, aircraft
problem complexity. The time for which each aircraft stays whose trajectories collide with the possible solution
in a cell depends on its model. It is assumed that each trajectories of the directly involved aircraft and whose
aircraft knows the list of cells that other aircraft or AV will velocities can be changed, and non-cooperative aircraft that
fly across. For example, in Figure 1 the list of cells of AV1 would maintain their initial velocities.
is: A5,B5, C5, D5, E5, F5, G5, H5, I5. DESCRIPTION OF CDR METHOD
In [19] a collision of two aircraft is predicted when they lie In this section the proposed CDR method for multiple
in the same cell of the discretized airspace at the same time. aircraft is presented. Firstly, the CDR method checks
However, two aircraft that are not in the same cell can be whether there is a collision among aircraft or not. If
closer than other two ones that are actually in the same cell collision is detected, it is solved by assigning a speed
55
London, UK, May 29-31, 2012 ATACCS’2012 | RESEARCH PAPERS
profile to each aircraft. Therefore, stay times of each in Figure 1, there are two arrival orders. The first one to be
aircraft in each cell could be changed. The resolution tested is AV2-AV1 because AV2 arrives before.
method, so-called Velocity Planning, consists of two steps: First, the tree is built for AV1, then for AV2 and so on
1. Search tree step: , which finds a solution if it when there are more AVs. Whenever a conflict zone is
exists by exploring all possible arrival orders to reached, the tree only calculates its following weight if all
each CZ AVs that precede the AV which is building its tree, given
2. QP-problem, optimization step which minimizes a by arrival order, have already calculated the weights of the
cost function edges related to that conflict zone (steps 8 and 9 in Figure
3). Figure 4 represents the complete trees in the proposed
example. Note that the horizontal length of each edge is
proportional to its weight. In this case, when AV1 comes in
conflict zone (t1) it does not continuous building its tree
because the AV before it, AV2, has not calculated its
weights of the edge related to that conflict zone, so the
AV1 tree is stopped and the AV2 tree is started and
completed. Otherwise the algorithm will continue with the
tree of the AV1.
56
London, UK, May 29-31, 2012 ATACCS’2012 | RESEARCH PAPERS
objective function can be done by solving a QP-problem.
The QP-problem minimizes a quadratic cost function with
linear constraints. The cost function is given by the
separation from the trajectory planned to fulfill the mission.
2
n mi
t t ref
J QP ik ref ik (3)
i 1 k 1 tik
where tik is the stay time in the kth zone visited by the ith AV
and tikref is the stay time in the initial trajectory. N
represents the number of AVs and mi the number of zones
crossed by the AV ith trajectory.
The stay time in each zone can be very different, so the
denominator has been included in order to minimize large
variations in the stay time in small zones. This would lead
to saturations in the velocity control signal. The constraints
Figure 4: Trees generated from the scenario showed in
of the model are:
Figure 1. AV2 tree only has one branch because is the first
AV that passes through of the conflict zone, CZ1. AV1 has tik aik vik bik 0
two branches because a collision has been detected in CZ1 cik vik d ik tik 0 (4)
so a backtracking process starts.
i 1...n, k 1...m
The complexity of the algorithm grows as the number of
conflict zones increases. Let us assume that there are m The maximum and minimum stay time in each cell depends
conflict zones with n AVs where ni of them are involved in on the initial speed in those cells, vik. This dependence is in
the ith conflict. So, a total of n1!...nm! different orders should fact nonlinear, but it can be linearized in order to formulate
be checked. the QP-problem. This is achieved by interpolation with aik,
bik, cik and dik as the interpolation coefficients. Moreover,
The first arrival order is determined by the estimated arrival the following constraints regarding vik are considered:
time to a conflict zone of each AV. Then, each tree is built
considering the arrival order to detect potential collision. If vik vmax 0
no solution is found with the first arrival order, a new order vmin vik 0
should be checked. The algorithm permutes the arrival (5)
ai ,max d i ,k 1 a d
order to the conflict zone from the cost function J. The vi ,k 1 vi ,k vi ,k 1 i ,max i ,k 1
vref vref
conflict zone with highest cost is chosen to define the
arrival order. This cost is defined as: i 1...n, k 1...m
J i i i (2) The first two constraints are given by the aircraft model and
the third one relates the initial speed in one cell with the
where μi and σi are the mean and standard deviation of the other in the previous cell because of the maximum
set of estimated arrival times of each AV to the conflict acceleration constraint. In the third equation, ai,max
zone. The mean of the arrival times to a conflict zone is represents the maximum desired acceleration of the ith
considered because a change in earlier conflict zones can aircraft and di,k represents the distance traveled by the ith
affect the following conflict zones. The standard deviation aircraft in the kth zone. Another constraint considered is the
is included in order to take into account the differences fulfillment of the Estimated Time of Arrival (ETA) at the
between the arrival times of the different AVs to a conflict end waypoint of each ith aircraft within a margin:
zone. In order to find a solution, it is advisable to change
mi
the arrival order to a conflict where all the estimated arrival
time of the vehicles involved in it are similar. t
j 1
ij t flight t m (6)
The backtracking process takes place when a collision is where tflight represents the initial estimated flight time, tm is
detected and a new tree should be built for one o more the maximum margin allowed, tij is the time spent in the jth
AVs. In this case, the backtracking process can become zone by the ith aircraft and mi is the number of zones
more complex than the previous case considering one crossed by the ith aircraft.
conflict zone. Note that the backtracking process is only
done for the AVs which arrive to the corresponding conflict Finally, for each conflict zone and each ith aircrfat that has
zone later. to cross that zone immediately before jth aircraft, the
following constraints should be considered in order to
QP-problem
avoid collisions:
When the search tree algorithm finds a solution, we have a
Q P 1
valid arrival order for all the conflict zones in the conflict
resolution problem. At this point, an improvement on the t
k 1
ik t jk 0
k 1
(7)
57
London, UK, May 29-31, 2012 ATACCS’2012 | RESEARCH PAPERS
th
where P indicates the entry cell in the CZ of the j aircrfat
and Q indicates the cell that has left ith aircraft.
The above optimization problem can be solved by means of
the QP-solver implemented in the Computational Geometry
Algorithms Library (CGAL) [21].
SIMULATIONS AND EXPERIMENTS
Simulations and experiments have been carried out to
validate the CDR method. The algorithms have been run in
a PC with a 2GHz Dual Core processor and 2 GB of RAM.
The operating system was Kubuntu Linux with kernel
2.6.32. The code has been written in the C++ language and
compiled with gcc-4.4.2.
A model of the aircraft is needed to simulate and evaluate
the suitability of the generated trajectories. Models of
different complexity can be used in the proposed method
[22][23]. The trade-off with the required computing time Figure 6: Second simulation scenario (S2).
should be considered. On the other hand, aircraft simulators TABLE I
can be used to generate the aircraft trajectories from real COMPUTATIONAL TIME SPENT AND CRITERIA RESULTS OF THE SOLUTIONS
flight data [24]. The simple model presented in [20] is used OBTAINED IN S1 FROM THE PROPOSED CDR METHOD
in the simulations. Two different scenarios are considered AVs QP(S) T(S) J(S2)
Conflict Detection(s) ST(s)
to perform the following two studies:
1. Analyze how the computing time depends on the 2 0.044 0 0.020 0.064 0.857
scenario considered (see Figure 5 and Figure 6). 3 0.188 0 0.096 0.284 0.859
4 0.716 0.004 0.708 1.428 0.861
2. Check how the computing time depends on the
safety distance or minimum separation. 5 1.804 0.008 2.608 4.420 0.894
6 4.032 0.024 11.425 15.481 0.896
7 7.456 0.028 29.854 37.332 0.967
8 14.065 0.052 83.717 97.834 0.969
9 23.073 0.064 179.455 202.593 1.105
10 38.078 0.104 287.046 325.228 1.109
58
London, UK, May 29-31, 2012 ATACCS’2012 | RESEARCH PAPERS
The computing time is lower in S2 because the trajectories of 1.5m. A conflict is detected in the center of the circle,
of each AV are shorter than in S1, so less time is spent in and then the speed profile is computed for each quad-rotor
simulation. to avoid the conflict in a cooperative way by minimizing
The computational cost of the proposed method when the the cost (3).
minimum separation (i.e., the number of cells) changes is
shown in Figure 7. Note that the computing time increases
as the minimum separation is increased. This extra time is
spent in the conflict detection phase of the algorithm and
arises due to the growth of conflict zones, which in turn
means that the conflict database contains more items and it
is more expensive to generate it.
59
London, UK, May 29-31, 2012 ATACCS’2012 | RESEARCH PAPERS
trajectory and fulfills its ETA at the end waypoint (6).
TABLE IV
These values are shown in bold in Table IV. SPEED PROFILE AND TIME FOR EACH AV IN EXPERIMENT II.
TABLE III AVs SP(m/s) T(s)
SPEED PROFILE AND TIME FOR EACH AV IN EXPERIMENT I.
1 0.736 3.689
AVs SP(m/s) T(s)
1 1.210 5.598
1 0.998 2.537 1 0.954 8.781
1 0.965 6.245 2 1.217 2.513
1 0.415 10.536 2 1.128 4.233
2 0.378 6.261 2 0.967 7.417
2 1.000 9.932 3 1.182 2.056
2 0.416 14.524 3 1.038 3.609
3 0.236 9.633 3 1.194 4.336
3 0.530 16.146 3 0.924 6.126
3 0.495 20.739 3 1.041 7.634
60
London, UK, May 29-31, 2012 ATACCS’2012 | RESEARCH PAPERS
resolution”, AIAA Journal of Guidance, Control and 16. A. Bicchi and L. Pallottino, “On optimal cooperative
Dynamics, vol. 25, no. 5, pp. 888-900, 2002. conflict resolution for air traffic management systems”,
8. L. Pallottino, E. M. Feron, and A. Bicchi, “Conflict IEEE Transactions on Intelligent Transportation
resolution problems for air traffic management systems Sysetms, vol. 1, no. 4, pp. 221-231, 2000.
solved with mixed integer programming”, IEEE 17. C. Tomlin, I. Mitchell, and R. Ghosh, “Safety
Transactions on Intelligent Transportation Systems, vol. verification of conflict resolution maneuvers”, IEEE
3, no. 1, pp. 3-11, 2002. Transactions on Intelligent Transportation Systems, vol.
9. R. Ehrmanntraut, “The potential of speed control”, in 2, no. 2, pp. 110-120, 2001.
Proceedings 23rd Digital Avionics Systems Conference, 18. A. Lecchini, W. Glover, J. Lygeros, and J. Maciejowski,
vol. 1, pp. 3.E.3-1-7, October 2004. “Monte carlo optimization for conflict resolution in air
10. E. Crück and J. Lygeros, “Subliminal air traffic control: traffic control”, IEEE Transactions on Intelligent
Human friendly control of a multi-agent system”, in Transportations Systems, vol. 7, no. 4, pp. 470-482,
Proceedings of the 2007 American Control Conference, 2006.
pp. 462-467, July 11-13 2007. 19. .J. J. Rebollo, A. Ollero and I. Maza, “Collision
11. A. Vela, S. Solak, W. Singhose, J.P. Clarke, “A Mixed avoidance among multiple aerial robots and other non-
Integer Program for Flight-Level Assignament and cooperative aircraft based on velocity planning,”
Speed Control for Conflict Resolution”, in Proceedings Proceedings of the 7th Conference on Mobile Robots,
48th IEEE Conference on Decision and Control, pp. Albufeira, Portugal, 2007.
5219-5226, 2009. 20. D. Alejo, R. Conde, J. Cobano, and A. Ollero, “Multi-
12. M. Christodoulou and S. Kodaxakis, “Automatic UAV collision avoidance with separation assurance
commercial aircraft collision avoidance in free flight: under uncertainties,” in Proceedings of the 5 th IEEE
The three dimensional problem”, IEEE Transactions on International Conference on Mechatronics (ICM 2009),
Intelligent Transportation Systems, vol. 7, no. 2, pp. pp. 1-6, Málaga (Spain), April 2009.
242-249, 2006. 21. CGAL, Computational Geometry Algorithms Library,
13. R. Vivona, D. Karr, and D. Roscoe, “Pattern Based http://www.cgal.org
genetic algorithm for airborne conflict resolution”, 22. I. Lymperopoulos, A. Lecchini, W. Glover, J.
AIAA Guidance, Navigation and Control Conference Maciejowski, and J. Lygeros, “A Stochastic Hybrid
and Exhibit, Keystone, Colorado, August 2006. Model for Air Traffic Management Processes,” Univ. of
14. R. Conde, D. Alejo, J.A. Cobano, A. Viguria and A. Cambridge, U.K., Technical Report CUED/F-
Ollero, “Conflict Detection and Resolution Method for INFENG/TR.572, Feb. 2007.
Cooperating Unmanned Aerial Vehicles ”, Journal of 23. BADA website:
Intelligent & Robotic Systems, DOI: 10.1007/s10846- http://www.eurocontrol.int/eec/public/standard_page/pr
011-9564-6, 2011. oj_BADA.html.
15. J. A. Cobano, R. Conde, D. Alejo and A. Ollero, “Path 24. N. Durand and J.M Alliot, “Optimization resolution of
Planning based on Genetic Algorithms and the Monte- en route conflicts”, First USA/Europe Air Traffic
Carlo method to avoid aerial vehicle collisions under Management Research and Development Seminar
uncertainties”, In Proceedings of the IEEE International (ATM1997), Saclay (France), 17-19 June 1997.
Conference on Robotics and Automation, Shaghai
(China), 2010.
61