You are on page 1of 6

Proceedings of the 45th IEEE Conference on Decision & Control WeIP3.

12
Manchester Grand Hyatt Hotel
San Diego, CA, USA, December 13-15, 2006

A Novel 3D Geometric Algorithm for Aircraft Autonomous


Collision Avoidance
C. Carbone, U. Ciniglio, F. Corraro and S. Luongo

Abstract-This paper presents a novel decision-making several minutes to compute the safe region with a simplified
algorithm for pair wise non-cooperative aircraft mid-air planar cinematic aircraft model having only three states. In
collision avoidance. An analytical solution for this control [6] an analytical solution is presented for the planar motion
problem, based on a three dimensional geometric approach, is of a single UAV by applying a geometric approach to the
derived. It does not require the solution of any programming
problem, thus resulting suitable for real-time applications. backwards reachable set computation.
Moreover, the availability of an analytical solution allows the Genetic algorithms and programming techniques have
application of well assessed control analysis and synthesis been also used for solving conflict resolution problems
techniques in order to improve stability and performance [7][8][9]. In these methods, conflict resolution is formulated
robustness. The proposed algorithm performs optimal as a constrained optimization problem, with the aim of
avoidance maneuvers, both in the horizontal and vertical plane, finding trajectories which minimize a proper cost function.
by minimizing aircraft deviation from its nominal trajectory. Its
effectiveness has been proved via numerical simulations, in The main drawback is that computation time is not
proper conflict scenarios which take into account aircraft predictable and the convergence to a solution is not ensured
dynamics and on-board sensors limitations. in a finite and deterministic time interval, as required by a
real-time control system.
I.INTRODUCTION In force field methods, each aircraft is treated as a charged
AIRCRAFT mid-air collision is still an unresolved particle and modified electrostatic equations are used to
problem, as available mishap data [1] show. This generate conflict resolution maneuvers [10]. The problem
situation is anticipated to become worse with the increasing with this approach is that the computed solutions are not
emerging traffic of small business aircraft, Very Light guaranteed to be feasible, once aircraft dynamic limitations
Aircraft (VLA) operating from and to secondary airports. On have been assigned.
the other hand, in order to increase aircraft capacity in the This paper presents a novel decision-making algorithm for
airspace, with the expected increase of civil Unmanned pair wise non-cooperative aircraft collision avoidance. Non-
Aircraft Vehicles (UAVs) a robust autonomous collision cooperative means that aircraft do not collaborate in
avoidance (ACA) system must be designed, developed and resolving the conflict, because either one of them is not
put in place [3]. Notice that an ACA system works on a equipped with sophisticated avionics, or communication
short-term time horizon (less than one minute) and it is between aircraft fails. In general, one aircraft is considered
considered as an emergency function autonomously engaged, as an intruder whereas the other one is assumed to be
at close ranges [2]. equipped with an ACA system, capable of detecting and
In general, it is composed of on-board detection sensors avoiding the intruder without knowing its intentions.
and decision-making algorithms. This paper focuses only on Furthermore, it shall be noted that the considered scenario
the decision-making algorithm part. does not involve an antagonist adversary at work, usually
A comprehensive survey of conflict detection and domain of game theory.
resolution approaches is provided in [4]. It is worthwhile The starting point in devising the algorithm of this paper
noticing that most of the methods presented in literature are has been a geometric approach (collision cone based) used in
not suitable for real-time applications, because of the non- planar mobile robotics [11] [12]. This approach has been
deterministic computational time needed for taking a extended to the 3D case and applied to aircraft. Usually, the
decision. In [5] authors solve the problem of two aircraft problem of designing an autonomous collision avoidance
conflict-prone as a two-person zero-sum dynamic game of algorithm is more critical when applied to aircraft, since they
the pursuer-evader variety, by computing reachable sets must maintain a minimum airspeed to guarantee a sufficient
defining regions of guaranteed safety. The main lift to remain aloft; moreover, lateral and vertical aircraft
disadvantage of this approach is that the algorithm takes dynamic limit maneuvering capabilities of the system.
Actually, an analytical geometric approach to the problem
of collision between two aircraft is also presented in [13], for
Manuscript received February 27, 2006. a planar scenario, and in [14] for a 3D environment using a
All authors are with the Flight Systems Department of CIRA (Italian
Aerospace Research Centre), Via Maiorise, 81043 Capua (CE), Italy. mixed geometric and collision cone approach; in this latter
Consider C. Carbone as corresponding author (phone: +39-0823- paper, analytical results are obtained for certain special cases
623309; fax: +39-0823-623521; e-mail: c.carbone@cira.it).

1-4244-0171-2/06/$20.00 ©2006 IEEE. 1 580


45th IEEE CDC, San Diego, USA, Dec. 13-15, 2006 WelP3.1 2

only, while numerical optimization techniques are employed two aircraft shall be detected, determining if their future
to search for solutions in the most general cases. positions, after a certain amount of time, experience a loss of
The algorithm presented in this paper, on the other hand, minimum separation (i.e., A/CA trajectory enters the safety
offers an analytical solution to the collision avoidance bubble surrounding A/CB); in such a case, the trajectory of
problem and does not require the solution of any numerical the aircraft with the ACA module on-board (A/CA) has to be
optimization problem, thus resulting very suitable for replanned by solving a conflict resolution problem.
real-time applications. This analytical solution opens the way
to the application of assessed non-linear control analysis and
VA =
(VA, VAY, VA ) = (VA cos A cos XA, VA cos rA sin XA, VA sin r ) V
synthesis techniques for a-priori performance and stability B
robustness evaluations. The algorithm effectiveness has been VB =
(VBX VBY, VBZ ) = (VB cos 7B cos XB, VB cos 7, sin XB, - VB sin 7, ) .xR
proved via numerical simulations, in proper conflict r =
(rx, ry, r,) = PB -PA
scenarios which take into account aircraft dynamic and on-
board sensors limitations.
The rest of the paper is organized as follows: the collision A>
avoidance problem is formulated as a two stage process: / NED~~~~XE
conflict detection and conflict resolution (Section II); °/ ~~~~NED
necessary and sufficient conditions for conflict detection are /~~~~~~~~~~~P XNED
stated in Section III, followed by the derivation of analytical
expressions for conflict resolution (Section IV); finally,
Section V presents numerical simulation results for the YNED
ZNED
validation of the collision avoidance decision-making Fig. 2. Collision geometry between a point of mass A and a sphere B
algorithm. with radius R.

II. PROBLEM FORMULATION III. CONFLICT DETECTION


Fig. 1 shows ACA module within the closed-loop control The relative velocity vector VAB =VA- VB transforms a
system. Its core is represented by a decision-making dynamic collision avoidance problem into a static problem:
algorithm, having as input speed and position of the intruder conflict geometry defined in Fig. 2 is equivalent to a
(1PB, VB ) and own aircraft (PA, VA). The outputs of the situation where sphere B is stationary and point of mass A
decision making algorithm are reference signals to the moves with relative velocity VAB (see Fig. 3).
autopilot, in terms of demanded speed module (Vd), slope
angle (Yd) and track angle (Xd).

,v Vd Yd

r=P -P

ACA system PA, VA VI


A

Fig. 1. ACA module within the closed-loop control system.

Conflict detection and resolution is addressed in a 3D r=P -P


environment using information of current positions and
instantaneous speed vectors.
Fig. 2 shows the geometry which has been adopted for a A VAB = VA VB
collision situation between two aircraft in the 3D North-East-
Down (NED) reference frame: the aircraft with ACA module Fig. 3. Definition of minimum separation distance vector dAB .
on-board (A/CA) is modeled as a point of mass - with a 3
Let dAB be a vector defined as the minimum separation
Degree Of Freedom (DOF) dynamic - having velocity VA,
whereas the other aircraft (A/CB, considered as an intruder) distance experienced between aircraft, after a certain time
is modeled as a sphere with radius R (safety bubble) having horizon. It can be calculated as follows (see Fig. 3):
velocity VB.
d = r V BV-
Given the above described geometry, a collision AB' ABV
2 VAB
(1)
avoidance can be formulated as a two stage problem: in the VAB

first stage, conflict detection, a potential conflict between

1581
45th IEEE CDC, San Diego, USA, Dec. 13-15, 2006 WelP3.1 2

Conflict detection stage is based on the following r


(Vrr+V0rF') rVVrVoV
AR
rj
d V2 'IVj2V
0~~~~~~~
rr= + 2
r V
r

Theorem 1: A point of mass A and a sphere B with radius


R which are moving in a 3D environment with constant
velocities, respectively VA and VB, are headed for a
AB
=r °V2
Finally,
collision if and only if the following conditions are satisfied: 2~ (
dAB <R2 # r2VV2<.2v+vR)r
2
dAB <R
V

dAB <R and r<0. (2)


Remark: Constant velocity vectors assumption in
theorem 1 is not restricting, since conflict detection
Proof: Let fTbe the plane on which lie vectors VAB and r .
conditions (2) can be evaluated continuously, with proper
This plane cuts sphere B determining a circle 6 having radius update rate l/Ts, according to the Nyquist-Shannon
R, as shown in Fig. 3. Consider this plane, highlighted also in sampling theorem. Consequently, reference signals to the
Fig. 4. Point of mass A and sphere B are headed for a autopilot are generated with sampling time Ts.
collision if and only if point of mass A - assumed to move
with velocity VAB in plane z - is headed for a collision with IV. CONFLICT RESOLUTION
circle 6, assumed to be stationary. With respect to versors r Consider the following collision cone definition:
and r', it is possible to write the following expressions:
%AB(R)-{VA(VA]7A*A) IVA ( El I2 I2
O, >d7YA
VAB =vgr+v6r0VO r
VAB =
Vr = r, Vo = rf. (3) (6)
r<Oj,dAB (VA )
r
*A] z,1r] < R}

According to theorem 1, all vectors VA belonging to the


Plane iv B collision cone CAB(R) are such that a collision is certain to
happen. This definition is purely geometric and does not take
into account aircraft dynamic limitations. The set of
avoidance solutions, compatible with aircraft dynamic
B~~~~A
limitations, can be computed as follows

A r
=f(Vd rdXd )E 31I Vd E [V v dE

(7)
d
,~~~~~~jl(V
*Ad E]-z,z], r>0 or dARA KA,Z
2'
A
R
Fig. 4. Plane zon which lie vectors VAB and F.

Notice that any solution of set Y works well, but the


In plane z the following theorem (proved in [11]) can be
tangential solution - which moves relative speed vector
applied:
If a point and a circle are moving with constant velocities VAB such that its projection line in time is tangent to the
such that their initial conditions satisfy sphere (see Fig. 3) - will result in an optimal conflict
r22RV < 2 Vr2 + V02) and Vr < O (4)
resolution maneuver. "Optimal" means that deviation from
the nominal trajectory is minimized, under the assumption of
then they are headedfor a collision. The above conditions straight trajectories and that a single step change in velocity
are both necessary and sufficientfor a collision to occur. vector can solve the conflict [13]. In order to find tangential
solutions it is possible to consider the following equation
Condition Vr = r<0 simply means that aircraft are 2
(|
|r _
AB)2
R2
approaching. In order to prove (2), it remains to prove the jAB
=
-~2 (8)
following equivalence VAB

dAB <
jAB <
R X r2V2
rV0 < R 2(V2 +V2) (5)
# Since in a conflict situation F VAB= |r|AB|cosj > 0,
At this aim, observe that expression (8) can be rewritten as follows
r =rr, VARVr+VBr
V r VA = rV
r VA =rVB +VAB F2 R2,
Hence, (9)
VAB= VVA +VB 2VAVBcos(fZVA, VB ).
1582
45th IEEE CDC, San Diego, USA, Dec. 13-15, 2006 WelP3.1 2

Considering vector components of VA, VBR and r, shown polynomial, as explained in appendix.
in Fig. 2, and equation (10),
COS(ZVA, VB ) VA VB = COS YA COS XA COS YB COS XB +
(0
VAVB
+ cos Asin XA cos YB sin XB+ sin YAsin rYB
After some trivial algebraic manipulations, (9) results in
the following equation

(rx cos Y cos A + ry cos YAsin *A- rz sin YA )2 +


2 VB (rx COS YA COS*A + ry COS YA sinA rz sin YA)
VA A.g
(1 1) A'/1
2 [VA + VB 2VA (VBX COS rA COS *A +
ff I,
Zfe .7

VB
A~~~~~~V

Equation (11) can be interpreted as follows: given F, VB Fig. 5. Backward and forward collision cones for lateral-directional
control.
and R, find all (VA,v,A ZA) corresponding to tangential
solutions in set Y. It can be solved numerically, but this is In a conflict situation, it can be proved that two real
computationally expensive for real-time applications. In solutions always exist (N.2). They correspond to the border
order to derive analytical solutions, this paper focuses on the of a forward collision cone, as shown in Fig. 5. When all
investigation of three different conflict resolution strategies, four solutions are real (N=4), a backward collision cone is
each one involving a single control variable: % (lateral- also present. In order to explain its meaning, consider a
directional control), y (longitudinal control) or V (speed conflict scenario in which intruder speed VB is greater than
control). It shall be noted that the optimality of the solution VA. In such a case, even though A/CA was able to turn back
is here considered with respect to the adopted single control instantaneously with its speed vector within the backward
command. collision cone, a collision would be inevitable.
Once the roots of (12) have been computed, the minimum
deviation of the aircraft speed vector - in terms of track
A. Lateral-Directional Control Strategy angle - with respect to the nominal trajectory, is calculated
In this control strategy, all avoidance maneuvers involving as follows
control variable %A are searched. In particular, VA and YA are AXd + min({ %A I1N) (13)
kept constant in (11), assuming as unknown only *A. In other
words, one searches for all solutions Xd such that velocity Let i be the index of the angle *' such that
vector VA is moved towards the collision cone border. By { x, -*AI N is minimized: Ad is chosen positive if
defining *i > ZA, negative otherwise.
jjq2 R2
Cy COSYA4 Ys = sinIYA 7H VA
K
VA
B
B. Longitudinal Control Strategy
equation (11) can be rearranged as follows This control strategy, on the other hand, searches for all
avoidance maneuvers involving control variable YA. In
al1 cos %A + a2,sin %A + a3% cOS %A s A+ (12) particular, VA and %A are kept constant in (11), assuming as
+ a4X coS *A+ a5sinf*A X=
bv unknown only yAd . By defining
where
a,, rXc, a2= ryc7,
= a3 =2r
CXr cCOS *A=,
Sin cAx + ry, ',=VBX CX+VBYSXv

a4X =2HVAVBXcY 2rxrZc7s7 2KrxC7, equation (11) can be rearranged as follows


a5, =2HVAVBYCY-2ry r cYsY- 2KrycY,
al. cos2 YA + a2, sin2 YA + a3, cos YA sin YA +
b,' H(V2AR)
=
HAVBZs7 r2s2
+V2)+2HVV -2Krzs - 2.
ZyZy. (14)
+ a4, cos YA + a5 sin YA by7,
Roots of (12) can be obtained by solving a fourth order where

1583
45th IEEE CDC, San Diego, USA, Dec. 13-15, 2006 WelP3.1 2

a,, =a
a2 = r2

a3, = -2arz
a4, =2(HVA/ - Ka)
a5y =2(Krz HVAVBZ)
by = H(V2+V2)- K2.
Also in this case, roots of (14) can be computed by solving
a quarter equation, as explained in appendix. Expression (13)
can be rewritten for the longitudinal control strategy, by Fig. 6. 3D test scenarios definition.
replacing % with and similar conclusions can be derived.
,

Fig. 7 shows the results of a set of simulations which have


V. NUMERICAL VALIDATION OF THE COLLISION AVOIDANCE been run starting from the proposed scenarios, adopting
ALGORITHM Lateral-directional Control. Each point of the figure
Algorithm validation has been carried out via numerical corresponds to a specific scenario and represents
simulations, by defining proper collision scenarios, as shown quantityS = min(r(t) - R), i.e. the minimum distance
in Fig. 6. These test scenarios have been designed according between A/CA trajectory and intruder safety bubble. S < 0
to standard specification given in [15]. It is assumed that means collision. As shown in Fig. 7, the proposed algorithm
onboard detection sensors can search from ±1 10° in azimuth has proved its effectiveness in all considered scenarios: S is
and ±15° in elevation. always positive. By adopting the same scenario, simulations
Conditions (2) also offer an analytical criterion to find have also been run using Longitudinal Control strategy.
initial positions of the two aircraft and their velocity vectors Results are summarized in Table I.
such that a collision situation is certain to happen, so to Table I
COMPARISON OF CONTROL STRATEGIES
drastically reduce the amount of collision geometries to be Control Strategy Number of Collisions
analyzed. Lateral-directional 0
In each conflict scenario it is assumed that intruder speed Longitudinal 5
vector is initially constant and such that a collision will Speed 135
certainly happen: Fig. 6 shows - for a given intruder (A/CB)
initial position - a "cone" of speed vectors that guarantee a Collisions met during Longitudinal Control strategy are
collision with A/CA, if a collision avoidance maneuver is not due to the aircraft dynamic limits, saturation constraints and
performed. a FOR limited to ±15° in elevation.
The proposed scenario foresees 2500 starting points Actually, a Speed Control strategy has also been
(r,P,O) for the intruder, chosen in a random way, with investigated for this scenario. For the same reasons explained
uniform distribution: in Longitudinal Control strategy, a large number of
r = 2.5 n mi, 'P U[- 10°, 1 10°], E) U[- 150, 15° ].
-
collisions has been encountered; in addition, there exist some
Simulations have been carried out under the following conflict geometries such that collision avoidance is
hypotheses: impossible by acting only on the speed module (e.g., a head
on conflict).
These results lead to the conclusion that Lateral-
i. Safety bubble radius: R = 152.4 m (500 ft) directional strategy is the best single-command control in the
ii. A/CA is of class VLA and modeled by a detailed considered scenario.
6DOF aircraft model
W. A/CA speed module constant: VA = 45 m/s VI. CONCLUSION
iv. A/CB speed module random: VB U[45 mls, 90 m/s] In this paper a novel decision-making algorithm for pair
V. After time tB U[5 s, 10 S], A/CB changes randomly wise non-cooperative aircraft collision avoidance has been
its track angle of I%B U[-20°, 20°]
-
presented. It has been derived on the basis of a three
Vi. On-board navigation sensors have been considered dimensional geometric approach, with the aim of using it for
with a realistic uncertainty model real-time applications.
vii. On-board detection sensors data-rate: 5 Hz The proposed algorithm offers an analytical solution to the
control problem of aircraft collision avoidance. This opens
Viii. On-board detection sensors can search from ±110° in the way to the application of assessed non-linear control
azimuth and ±15° in elevation (Field Of Regard, analysis and synthesis techniques for a-priori stability
FOR) [2]. robustness evaluations and performance improvement. In this

1584
45th IEEE CDC, San Diego, USA, Dec. 13-15, 2006 WelP3.1 2

paper three different conflict resolution strategies have been P4 a, -a4- b, p3 = 2(a5 -a3) P2 = 2(2a2 -a1 - b)
investigated, each one involving a single control variable - * P1 2(a3+ a5 ), po a, +a4 -b.
(lateral-directional control), y (longitudinal control) or V
(speed control) - and a comparison among these strategies There exists a classical method (due to L. Ferrari, in 1540)
has been carried out. The effectiveness of the proposed for calculating analytically the roots of a quartic equation
approach has been proved via numerical simulations, in (here omitted for the sake of brevity).
proper conflict scenarios which take into account aircraft Note that tg (x/2) can be inverted if
dynamics and on-board sensors limitations. The algorithm
has proved its validity in all considered scenarios, -z/2<x/2<z/2, 9-z<x<z; hence, it can always be
performing optimal avoidance maneuvers. Real-time inverted for ZA and YA.
hardware in the loop simulations and flight tests on a VLA
flying test bed have been scheduled and are in progress. ACKNOWLEDGEMENT
Currently, a new geometric algorithm - which uses all This work was supported by TECVOL, an internal CIRA
control variables at the same time (track, slope and speed) - project, and IFATS (Innovative Future Air Transport
is under development. The analytical tangential solution is System) project, under contract No. AST3-CT-2004-
searched directly on the 7t-plane, yielding a 3D optimal 503019-IFATS.
collision avoidance maneuver.
Moreover, an extension of this approach to REFERENCES
multiple-aircraft is being researching. [1] FAA Aviation Safety Data, FAA website
[2] W.Z. Chen, T.J. Molnar, "Autonomous Hight Control Sensing
Lateral-Directional Control Strategy N19 :0 Technologies - Capability Goals and Sensing Requirements", AFRL-
collsion VA-WP-TP-2002-309, 2002.
r =d. *1 ... =i= .4. [3] 25 Nations for an Aerospace Breakthrough - "European Civil UAV
Roadmap" - Final Report, Thematic Network on European Civil UAV
I uu .±..~..................... .. ..........
[4]
FP5 R&D, 2005.
J. K. Kuchar and L. C. Yang, "A Review of Conflict Detection and
Resolution Modelling Methods", IEEE Transactions on Intelligent
0 Transportation Systems, vol. 1, No. 4, December 2000.
[5] C. Tomlin, G. J. Pappas, and S. Sastry, "Conflict Resolution for Air
0 Traffic Management: A Study in Multiagent Hybrid Systems," IEEE
Transactions on Automatic Control, Vol. 43, pp. 509-521, April
E -50 -_ 1998.
[6] E. Frew and R. Sengupta, "Obstacle Avoidance with Sensor
M
-166l Uncertainty for Small Unmanned Aircraft", 43rd IEEE Conference on
Decision and Control, Bahamas, December, 2004.
[7] L. Pallottino, E. M. Feron and A. Bicchi, "Conflict Resolution
Problems for Air Traffic Management Systems Solved with Mixed
5 W0 1666 1566 2000 Integer Programming", IEEE Transactions on Intelligent
Scenarios Transportation Systems, Vol. 3, No. 1, March 2002.
Fig. 7. Lateral-directional control testing: each point corresponds to [8] J.-M. Alliot, H. Gruber, G. Joly and M. Schoenauer, "Genetic
a specific scenario and represents S = min( r(t) - R) [m]. Algorithms for Solving Air Traffic Control Conflicts", Artificial
Intelligence for Applications, 1993. Proceedings, Ninth Conference
on, 1-5 March 1993, Pages: 338-344.
APPENDIX [9] E. Frazzoli, Z.-H. Mao, J.-H. Oh, and E. Feron, "Resolution of
In order to solve (12) and (14), consider the following Conflicts Involving Many Aircraft via Semidefinite Programming",
Journal of Guidance, Control, and Dynamics, Vol. 24, No. 1,
generic equation January-February 2001.
[10] R. Ghosh and C. Tomlin, "Maneuver design for multiple aircraft
a cos2 x+a2 sin x+a3 cosxsinx+a4 cosx+a5 sinx =b. conflict resolution", American Control Conference. Proceedings of
the 2000, Vol. 1, Issue: 6, 28-30 June 2000, Pages: 672-676.
[11] A. Chakravarthy and D. Ghose, "Obstacle Avoidance in a Dynamic
Substituting Environment: A Collision Cone Approach", IEEE Transactions on
Systems, Man and Cybernetics - Part A: Systems and Humans, Vol.
I-t2
28, No. 5, September 1998.
x
2t2
sin X = 7 COS X: : t2 '
t = tg2
~2 [12] P. Fiorini and Z. Shiller, "Motion Planning in Dynamic Environments
using Velocity Obstacles", Int. Journal of Robotics Research, Vol.17,
and simplifying, the following quartic equation is derived No.7, pp.760-772, July 1998.
[13] K.-D. Bilimoria, "A Geometric Optimization Approach to Aircraft
Conflict Resolution", AIAA Guidance, Navigation, and Control
p4t4 + p3t3 + p2t2 + plt + Po =0. (15) Conference, August 2000, Denver, Colorado.
[14] J. Gross, R. Rajvanshi and K. Subbarao, "Aircraft Conflict Detection
and Resolution using Mixed Geometric and Collision Cone
Where, its coefficients are as follows, Approaches", AIAA Guidance, Navigation, and Control Conference
and Exhibit, Providence, Rhode Island, 2004.
[15] "Standard Specification for Design and Performance of an Airborne
Sense-and-Avoid System (F2411-04)", ASTM International, 2004.

1585

You might also like