3 SafetyAssesment Self
3 SafetyAssesment Self
3 SafetyAssesment Self
DOI
10.2514/6.2016-1000
Publication date
2016
Document Version
Peer reviewed version
Published in
Proceedings of the AIAA infotech at aerospace
Citation (APA)
Yazdi Ibrahim Jenie, Y., van Kampen, E. J., Ellerbroek, J., & Hoekstra, J. M. (2016). Safety assessment of
unmanned aerial vehicle operations in an integrated airspace. In s.n. (Ed.), Proceedings of the AIAA
infotech at aerospace (pp. 1-11). Reston: AIAA. DOI: 10.2514/6.2016-1000
Important note
To cite this publication, please use the final published version (if applicable).
Please check the document version above.
Copyright
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent
of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Takedown policy
Please contact us and provide details if you believe this document breaches copyrights.
We will remove access to the work immediately and investigate your claim.
This paper focuses on the safety assessment of Unmanned Aerial Vehicles (UAVs) op-
erating in an integrated airspace system. The assessment is based on series of Monte
Carlo Simulations to include the effect of the variety of Conflict Detection and Resolution
(CD&R) systems in each involved vehicle. The difficulty of using Monte Carlo simulations
in the assessment, in which collision are rare to be sufficiently represented, is overcome by
the use of a high density airspace with a periodic boundary condition. This setup, along
with the randomization of the UAV states and the CD&R parameters, increases the num-
ber of conflicts and resolutions for each Monte Carlo sample to rapidly reach convergence.
The parameters derived, including the mean ratio of time in which the vehicle are in either
mission mode, avoidance mode, or in a Near Mid Air Collision (NMAC) situation, are
compared to similar Monte Carlo simulation results where no CD&R system is used. The
proposed method is a versatile safety assessment method for various encounter situations
a UAV an its CD&R system might face in its operation in an integrated airspace system.
Nomenclature
I. Introduction
ntegrating Unmanned Aerial Vehicles (UAVs) into the current air traffic presents new challenges for the
Idevelopers,
airspace management, especially in ensuring their operational safety. With the potentially large number of
operators, and missions, the vast variation of methods in UAV Conflict Detection and Resolution
(CD&R) system is inevitable, which would create a heterogeneous environment for operation. This becomes
a problem since most of the CD&R systems are yet to be proven in handling such heterogeneous situations, in
the integrated airspace. On the other hand, subjected to a standardized CD&R procedures in manned-flight,
both airspace authorities and regulations cannot yet accommodate the heterogeneousness as well. Taking
safe measures, current policy to enable the integration of UAVs in the airspace is to define a target level of
safety (TLOS), as it is stated in 1. One parameter of the target level of safety is the probability of Near
Mid-Air Collision (NMAC). If strictly taken, the TLOS for NMAC probability for a UAV in an airspace
should be less than one collision out of a billion operations (10−9 ).
While many methods have been proposed to analytically asses the operational level of safety, most of
them do not include the effects of the CD&R system in their calculation, let alone a variation in the system.
Ref. 2 and 3, for instance, introduce the derivation of risk of collision as the expected number of collisions
over a specified period of time, assuming constant velocities of the vehicle. Using the same assumption, Ref.
4 introduce a gas model for two and three body representations, making calculations based on a homogeneous
distribution of aircraft in several different density assumptions. Most of these studies define uncertainties in
the vehicle positions and velocities in the NMAC probability derivations.
Including the CD&R factor in a safety assessment is in fact difficult without an extensive method of
modeling and simulation of the vehicle dynamics, such as by performing Monte Carlo simulations,5 or by
using dynamic programming.6 With these methods, random samples can be inserted to model the variation
of the UAV CD&R system and initial states in an airspace. Repeated simulation is then conducted across
the samples that generates various conflict encounters and resolutions, and in the end draws out required
parameters for safety assessment. These methods, however, are rarely desirable for safety analysis, especially
when estimating the probability of collisions, since, in a realistic airspace density, collisions are very rare
that it will take a considerably large amount of time and samples to obtain a sufficient result. Most of the
time, the samples will result in zero collision occurrence, such as presented in 7 and 8 .
The current paper presents a method of safety assessment for UAV operations in an integrated airspace by
resorting to the Monte Carlo simulations. To overcome the drawbacks of the method, a high density airspace
is modeled, where a hundred UAVs are initiated, positioned close together in an upright square-lattice with
random headings and speeds. The heterogeneousness of the air traffic is modeled in two elements, i.e., the
uncertainties of detection and the variation of conflict resolutions. The density of the airspace is maintained
using a Periodic Boundary Condition. This setup wraps the movement of vehicles inside the area of interest,
eliminating the unavailing samples of the Monte Carlo simulations, such as, for example, vehicles that leave
the area of interest before having chance to encounter a conflict.
The simulation setup will make the Monte Carlo process reaches convergence faster with the increase
the chances for each vehicle to encounter, and therefore resolve, conflicts. Furthermore, since the safety
parameter is averaged for the vehicles, each of the airspace samples already have one hundred samples of
random conflicts and avoidances, which in the end speeds up the parameter convergence even more.
This paper is structured as follows. After this introduction, Section II discusses the heterogeneous air
traffic model, which consists of three main elements, i.e. (1) the high density airspace, (2) the uncertainties
of detection, and (3) the variety of resolutions. Section III presents the result of the Monte Carlo simu-
lation based on that model. Results of the simulations where no CD&R is involved are also presented for
comparison, along with a brief discussion.Section V concludes the paper and provides suggestions for future
work.
of the CD&R architecture. The 3D part, however, is a case that is less stressful than the 2-Dimension, since
the vehicles can exploit more space in order to reach safety.
Three key elements of the simulator, to represent the work of heterogeneous CD&R system in an inte-
grated airspace, are explained in this section, which includes the airspace representation, the uncertainty in
detection, the variety of resolution. These elements are also the source of collisions and is designed to produce
as many collisions as possible in order to reduce the Monte Carlo sample requirement while maintaining the
rationale. Table 1 lists the parameter randomization for the Monte Carlo simulations, which is explained in
the following subsections.
1
σx = −1, if (x(k) + vx (k)∆t) > 2 Lx
σy = −1,
if (y(k) + vy (k)∆t) > 12 Ly
σx = 1, if (x(k) + vx (k)∆t) < − 12 Lx , σy = 1, if (y(k) + vy (k)∆t) < − 12 Ly (2)
σx = 0, otherwise σy = 0, otherwise
(a) (b)
Figure 1. (a) initial setup of the airspace with 100 vehicles arranged in a square lattice; (b) The Wrap-around
effect of the Periodic Boundary Condition applied in the airspace
In implementing the PBC, the wrap-around effect need to be considered in the conflict detection and
resolution generation. This is illustrate in Figure 1-b, where vehicle B need to avoid C, which position is
actually far away from B on the upper left corner of the area. Therefore, the ’clones’ of vehicles near to the
edges should be considered in the Velocity Obstacle resolutions.
In cases where body-to-body collision occurred due to failure in avoidance, the density of the area cannot
be maintaine, since all involved vehicles are removed immediately. Although this setup might make some of
the Monte Carlo simulation samples unavailing, the chances of body-to-body collision should be very small
if the heterogeneous CD&R performs well. Loss of separations, on the other hand, does not remove the
involved vehicles from the simulation, and therefore the vehicles might face loss of separation more than
once.
" # " #
~ i∗ = x∗i x i + x
D = (3)
yi∗ yi + x
" # " #
~∗ = vi∗x vi x + v
V i = (4)
vi∗y vi y + v
h i
V~o − V
~∗ ·D ~∗
i i
~o ∈ VOi ⇐⇒ 0 <
V < cos αvo ∧ D ~ ∗ < davo (7)
~ i
Vo − V~ ∗ D
~ ∗
i i
The avoidance velocity vectors Vavo = vavox , vavoy for both direction can be derived geometrically, by
the intersection of the ownship reachable velocity circle with both edge of the VO. There will be four set
of solutions that should satisfied system of linear and quadratic equation presented in equation (8) to (10),
which are functions of davo and the radius of the separation zone rsep , in a particular obstacle bearing (ψoi )
~o and V
and velocity of the vehicles (V ~i ).
Figure 2. The Velocity Obstacle method for an example of two-dimensional encounter case
2
~
va2y + va2x = Vo (9)
where the coefficient mvo in equation (8) represents the two edges of the VO set, presented in equation
(10) for both left and right turning directions. To ensure the existence of solutions in a circumstance when
the avoidance have to be conducted inside the separation zone, the term d2avo − rsep 2
is set to be zero
whenever davo < rsep . This particular case might occur when a avoidance failed due to dynamic interaction
of the vehicles. Hence the resolution changes from separation maintaining to separation regaining, to direct
the vehicle to leave the separation zone as fast as possible.
q
δχ x∗i rsep + yi∗ d2avo − rsep
2 −d ∗
avo viy
(
δχ = 1 , for left turning
mvo = − q , (10)
∗ 2 2 ∗ ∗
xi davo − rsep − δχ yi rsep − davo vix δ χ = −1 , for right turning
The Vavo for both direction are chosen from the set of Va solution that have at most two real members,
which are the two closest solution from the current Vo , satisfying the argument in equation (11).
n o
Va | vax , vay ∈ R, vay > 0, min vay
vax , for left turning
Vavo = n vax o (11)
va y va y
Va | vax , vay ∈ R,
va < 0, min − va
x x
, for right turning
The variation of resolution is produced by alternating the distance where the avoidance started (davo ), and
the radius of the separation zone (rsep ). The former parameter affects the aggressiveness of the resolution:
the closer the avoidance starting point from the obstacle, the more aggressive the maneuver will be. On the
other hand, the radius of the separation zone, under the state diagram in Figure 2, is exactly the closest
point of approach (CPA) between the vehicles. Example of conflict resolution variation can be observed in
Figure 3, which is simulated with assumption of instantaneous change of velocity in one time step.
Hence, the parameters davo and rsep are uniformly randomized between a specified range to create a
heterogeneous situation in an integrated airspace with variety of resolution maneuver. The avoidance starts
between 10 and 100 meters, which direct the vehicles to avoid other with CPA between 5 and 25 meters.
(a) (b)
Figure 3. Example of resolutions varieties due to different preferences of (a) avoidance starting distance (davo )
and (b) Separation radius (rsep )
(a)
(b)
(c)
Figure 4. Example of the High-Density airspace simulation for three different airspace density
Nv mode
mode
1 X ti
τ̄j = (12)
Nv i=1 Tj
The expected value of those time parameters is therefore approximated using their averages across the
Monte Carlo airspace samples-j = 1, 2, ..., NM C , as it is presented in equation (13). Notice that these
parameters, τ̄ mode , have also already been averaged in equation (12) across the 100 vehicles in each sample.
The parameter variances are estimated in the Monte Carlo simulation, using equation
N MC
1 X
E [τ̄ mode ] = τ̄jmode (13)
NM C j=1
NMC
1 X 2
στ̄2 mode =
τ̄jmode − E [τ̄ mode ] (14)
NM C − 1 j=1
Finally, the precision of the Monte Carlo estimation is concluded using the central limit theory, resulting
the parameter to be within the range presented in equation (15), for 99.95% confidence interval.
3.3 στ̄ | 3.3 στ̄ |
τ̄ mode = E [τ̄ mode ] − √ mode , E [τ̄ mode ] + √ mode (15)
NM C NM C
The overall result of the time parameter derivation is listed in Table 2. An example of the parameters
derivation in shown in the τ̄NMAC convergence graph in Figure 5. The parameter converges quickly after
approximately 50 airspace samples, demonstrate the effectiveness of the simulation setup in generating
sufficient samples of random conflicts encounters and resolutions. The shaded background on each plot in
Figure 5 represent the corresponding 99.95% confidence interval.
Figure 5 compares cases where no avoidance implemented and cases where CD&R is activated. The
result for the first case is basically the chance of NMAC occurrence in the airspace. The result shows that
the CD&R implementation are able to reduce the NMAC occurrence, however not to the level expected,
especially on the cases with 25/km2 density that leaves more than 2.61 ± 0.30% chance of NMAC. Even in
the least dense airspace of the simulations, the use of CD&R system cannot guaranteed the safety, while
they actually demonstrate a reduction from 7.35 ± 0.54% to only 0.30 ± 0.04% chance of NMAC.
Figure 5. Monte Carlo simulation results for the Mean Time of NMAC, for case without and with CD&R
(inset)
Figure 7 shows the three mean-time parameters, on operations with the CD&R in a heterogeneous
airspace, in a bar graph for easier comparisons. Evidently, the mean-time when the UAVs are able to
perform their mission increases as the density of the airspace decreases. In the highest density of 25 vehicle
per kilometer square, the time to perform the mission is almost the same with the time to avoid. In this
high density setup, the results even shows that from the avoidance performed, most of the time the vehicle
is in avoidance mode, meaning that it change its direction, trying to find a resolution path, around 30% of
the time. this is also have been shown before, that on this density, the algorithm actually does not find a
resolution and instead, the conflict fade by it self due to the direction difference. Other density setup shows
the domination of maintain mode, in which the vehicle is directed to a specific resolution path.
Table 3 shows another simulation setup, with a level of maneuver coordination by forcing every vehicle to
avoid by turning to the right. This setup intended to eliminates the problem of reciprocating dance, where
Figure 6. Mdododo
Figure 7. Mean-time parameters results for operation with CD&R in heterogeneous airspace, in three different
density
a pair of vehicle avoid each other to the same side, which caused a new conflict. Coordinating by turning to
the right is also a part of the Visual Flight Rule (VFR)13 in manned flight.
The table, however, shows an adverse result. On the high density airspace, the NMAC risk actually
gets much higher, to 7.06 ± 0.53%. On the less dense airspace, however, an insignificant reduction of the
NMAC risk is produced. These probably happen because while the coordination eliminates the reciprocating
dance on head-on encounter, it produce more dance in the taking-over maneuver, where vehicles are moving
approximately in the same direction. Randomization of avoidance direction is actually benefiting, since
it lower the chance of reciprocating dance in both head-on and taking over encounter, and opening more
avoidance path, especially in a high density airspace.
Mean-time of mission in the coordinated airspace, however, is recorded significantly higher for all three
different density. This is due to the less time it needed for each avoidance to clear the conflict, since both
vehicle are avoiding in the contributing direction.
IV. Conclusion
This paper has assessed the safety of Unmanned Aerial Vehicles (UAVs) operation when it is integrated
in the airspace system. Three parameters have been derived including the mean ratio of time in which the
vehicle are in either mission mode, avoidance mode, or in a Near Mid Air Collision (NMAC) situation for
three case of airspace density. It is demonstrated that Monte Carlo simulation can be used to asses the safety
by simulation setup of high density airspace, coupled with the the use of periodic boundary conditions.
While the method is able to assess the safety in various operation condition in an integrated airspace,
several limitations are noted. The method has not been demonstrated in a Three-dimensional setup, while
the two-dimensional is more stressful. Nevertheless, the proposed method is a versatile safety assessment
method for various encounter situations a UAV might face in its operation in an integrated airspace system.
References
1 Dalamagkidis, K., Valavanis, K. P., and Piegl, L. A., “On Integrating Unmanned Aircraft Systems into the National
Airspace System,” International Series on Intelligent Systems, Control, and Automation: Science and Engineering, Vol. 36,
Springer Science+Business Media, 1st ed., 2009, doi: 10.1007/978-94-007-2479-2.
2 Reich, P. G., “Analysis of Long-Range Air Traffic Systems: Separation Standards I,” The Journal of Navigation, Vol. 50,
doi:10.1109/PROC.1970.7643.
5 Jenie, Y. I., van Kampen, E.-J., and Remes, B., “Cooperative Autonomous Collision Avoidance System for Unmanned
Aerial Vehicle,” Advances in Aerospace Guidance, Navigation and Control, edited by Q. Chu, B. Mulder, D. Choukroun, E.-J.
Kampen, C. Visser, and G. Looye, Springer Berlin Heidelberg, 2013, pp. 387–405, doi: 10.1007/978-3-642-38253-6 24.
6 Kochenderfer, M. J., Holland, J. E., and Chryssanthacopoulos, J. P., “Next Generation Airborne Collision Avoidance
System,” Lincoln Laboratory Journal, Vol. 19, No. 1, 2012, pp. 55–71, doi: 10.2747/1548-1603.48.1.24.
7 Jenie, Y. I., van Kampen, E., de Visser, C. C., and Chu, Q. P., “Selective Velocity Obstacle Method for Cooperative
Autonomous Collision Avoidance System for UAVs,” AIAA Guidance, Navigation, and Control Conference 2013 , AIAA,
Boston, MA, 2013.
8 Jenie, Y. I., Kampen, E.-J. v., de Visser, C. C., Ellerbroek, J., and Hoekstra, J. M., “Selective Velocity Obstacle Method
for Deconflicting Maneuvers Applied to Unmanned Aerial Vehicles,” Journal of Guidance, Control, and Dynamics, Vol. 38,
No. 6, 2015, pp. 1140–1146.
9 Mohleji, S. C. and Wang, G., “Modeling ADS-B Position and Velocity Errors for Airborne Merging and Spacing in
Systems, IEEE Transactions on, Vol. 1, No. 4, Dec 2000, pp. 179–189, doi:10.1109/6979.898217.
11 Jenie, Y. I., van Kampen, E., Ellerbroek, J., and Hoekstra, J. M., “Conflict Detection and Resolution System Architec-
ture for Unmanned Aerial Vehicles in Civil Airspace,” AIAA Infotech @ Aerospace, AIAA Scitech Conference 2015 , AIAA,
Kissimmee, FL, 2015, AIAA 2015-0483, doi:10.2514/6.2015-0483.
12 Fiorini, P. and Shiller, Z., “Motion Planning in Dynamic Environments Using Velocity Obstacles,” The International
Journal of Robotics Research, Vol. 17, No. 7, 1998, pp. 760–772, doi: 10.1177/027836499801700706.
13 Federal Aviation Administration, Federal Aviation Regulations (FAR) Chapter I, subchapter F Air Traffic and General