Original Paper

Measurement and Control

1–8
Robust path planning for avoiding Ó The Author(s) 2019
Article reuse guidelines:
obstacles using time-environment sagepub.com/journals-permissions
DOI: 10.1177/0020294019847704

dynamic map journals.sagepub.com/home/mac

Xiaolin Zhao1 , Yu Zhang2 and Boxin Zhao1

Abstract
Small unmanned aerial vehicles are widely used in urban space because of its flexibility and maneuverability. However,
there are full of dynamic obstacles and immobile obstacles which will affect safe flying in urban space. In this paper, a
novel integrated path planning approach for unmanned aerial vehicles is presented, which is consisted of three steps.
First, a time-environment dynamic map is constructed to represent obstacles by introducing time axis. Second,
unmanned aerial vehicles’ flyable paths are explored based on breadth-first algorithm. Third, a path planning method
using A* algorithm and local trace-back model is designed in order to discover sub-optimal feasible path rapidly in
unmanned aerial vehicles’ field of view. Finally, the simulation results have illustrated that the proposed method can
ensure unmanned aerial vehicles’ autonomous path planning safely and effectively in urban space crowded with
obstacles.

Keywords
Path planning, unmanned aerial vehicles, time-environment dynamic map

Date received: 31 January 2019; accepted: 22 March 2019

Introduction flying, so that a safe path from start-point to end-point

can be designed according to the collected environmen-
Since unmanned aerial vehicles (UAVs) have great tal information. However, it is inaccessible to obtain
potential to complete missions without human inter- reliable information of the entire environment, for
vention,1 large number of applications of UAVs for instance, pop-up obstacles will never be localized in
both military and commercial purposes have been advance until a UAV is approaching them.8 Thus, the
emerged. However, in the urban space, the UAVs usu- traditional graph-based approaches can hardly meet
ally encounter with various obstacles, such as buildings, the requirements of planning a safe path in dynamic
electrical wires and other flying UAVs, which may
environment.
affect UAVs’ safety. Therefore, designing a safe as well It turns out that UAVs must have the ability to deal
as flyable path for UAVs in the environment crowded
with new obstacles that burst into environments.
with obstacles is needed.
Furthermore, there are many works in the literature
In recent years, many approaches about path plan-
dealing with path planning for UAVs in dynamic envir-
ning have been investigated for UAVs. McLain and
onments, such as A* and D* algorithms,9–11 potential
Beard2 constructed a Voronoi diagram based on the
fields,12 model predictive control13 and evolutionary
known locations of threats to generate an initial path
algorithmic techniques.14,15 Many efforts have been
and then took different combinations of the edges so as
made in the field of UAV path planning, and these
to choose the lowest-cost flyable path. As pointed out
approaches can be mainly classified in two thoughts:
by Wang et al.,3 this approach did not take the threats’
effective range into consideration, and thus a UAV
may go through some of the threat zones. Wang et al.
developed a path planning method, which built a 1
Air Force Engineering University, Xi’an, China
2
Laguerre diagram based on Delaunay graphs, to over- Department of Electronic Engineering, Tsinghua University, Beijing,
come the drawbacks of Voronoi diagram. The graph- China
based methods such as Voronoi diagrams,2,4,5 Laguerre Corresponding author:
diagram3,6 or visibility graphs7 all assume that the Boxin Zhao, Air Force Engineering University, Xi’an 710038, China.
entire environment remains unchanged when a UAV is Email: [email protected]

Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License
(http://www.creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 Measurement and Control

1. Assuming that the number of obstacles is lim-

ited and obstacles are far from each other so
that a safe path can be easily designed.2–7 This
assumption is not suitable for urban environ-
ments because irregular obstacles are densely
placed in the environment. Thus, flyable areas
are limited.
2. Calculating the obstacles’ probability of occur-
rence at some points in the future via dynamic
prediction algorithms so as to plan a path with-
out obstacles for UAVs.16,17 However, these
methods highly depend on the accuracy of loca- Figure 1. Sketch map of urban space crowded with obstacles.
tion and the efficiency of algorithms, and thus
are not practical for real-time path planning in
the dynamically changing environments.

In this paper, we propose a collision avoidance algo-

rithm focusing on dense and static threats in uncertain
environments.18 In order to plan a safe flyable path in
urban space crowded with irregular buildings and mov-
ing obstacles, we introduce a time axis to the collision
avoidance algorithm. In this manner, the environments
where UAVs fly are extended to time-environment
dynamic map (TEDM). Based on TEDM, we designed
an integrated approach which combines breadth-first
algorithm, A* algorithm and local trace-back model to
plan an optimal path in each short time interval, thus
avoiding irregular buildings without simplification. Our
proposed method is detailed in the following sections.
Figure 2. The kinematic model of UAV.

Problem statements and UAV modeling

UAV modeling
Problem statement The results we present in this paper can be applied to
With popularity and proven success of UAVs in vari- different kinds of rotorcrafts. In this paper, we focus
ous situations, the demand of UAVs used in urban on quadrotor UAVs for it is an intuitive model that is
space is increasing. The remarkable character of urban often used in literature.
space is that, as we all know, it is filled with obstacles The key features of quadrotor aircrafts lie in its flex-
which are irregular and hazardous, and the flyable ibility and maneuverability so it comes out that we do
region between buildings is restricted. To avoid flight- not need to think much about the restrictions of the
risks of obstacle collision, we design a novel path minimum turning radius and the overload problem
planning algorithm for UAVs in urban space crowded while turning. As a result, a four-rotor aircraft can be
with irregular obstacles, which makes use of the cost regarded as a controllable particle,19 without consider-
function to build an optimal local trace-back model ing the rotation around center of mass. So, the kine-
within the range of sensors. The dynamic obstacles in matic model of UAV is defined by its position
environment should also be considered, which is an !
(xuav , yuav ), orientation u and velocity vector V uav (see
enormous challenge for path planning in such a strict Figure 2).
circumstance. Figure 1 depicts a representative urban Therefore, the UAV’s kinematic equation is given by
space, in which the gray boxes indicate building obsta-
(
cles, the black spots mean other dynamic UAVs which !
x_ uav = V uav cos u
represent moving obstacles and the arrows mean the ! ð1Þ
y_ uav = V uav sin u
corresponding direction. Given such an environment,
our mission is to plan a safe path for a UAV from the Note that Vmin 40 and that a negative velocity corre-
start position to the final position, which passes sponds to moving backward. Because we will apply the
through the path from the target position 1 to the tar- UAV gridding model later, it is convenient to choose
get position 2 sequentially. The red spots in Figure 1 the symmetric bounds for the velocity term.
stand for the target positions, respectively. For convenience, many works assume that the
UAVs maintain constant velocity;20,21 however, it is
Zhao et al. 3

Figure 3. Gridding model of UAV.

Figure 5. Horizontal plane of urban space.

to replace three-dimensional space while constructing

the model of static obstacles (see the gray regions in
Figure 5).
To moving obstacles (see Figure 5, the black area),
time axis is introduced on the basis of the horizontal
plane we built before to discretize the environment,
forming a three-dimensional TEDM. It consists of state
hx, y, v, u, ti, where D obstaclei = hxi , yi , vi , ui , ti con-
Figure 4. Sketch map of FOV. tains a finite number of dynamic obstacles with
i = 1, 2, ., N, specific time point t, the ith dynamic
obstacle’s position (xi , yi ), its velocity vi and orientation
still not easy for UAV’s controller to compute because ui . Set the update rate of TEDM as DtTEDM , and for
constant variables are often inestimable. Hence, the the dynamic obstacles
UAV’s kinematic equation based on particle is further
simplified using gridding model.18 As one can observe lim ðvt + DtTEDM  vt , ut + DtTEDM  ut Þ = ð0, 0Þ ð3Þ
DtTEDM !0
from Figure 3, a gridding model of UAV is able to
move in eight geographical directions in case they are which means vi and ui remain unchanged in the short
free. time period ½t, t + DtTEDM  where DtTEDM is infinitesi-
Taking the factor into consideration that UAV needs mal; here, we define DtTEDM equals to the quotient
to explore obstacles around to get information about value between the velocity of the quadrotor and the
the environment, we define the boundary of detected length from the UAV to the edge of FOV. It can be
coverage as field of view (FOV, which is similar to the seen that if we locate the dynamic obstacles in UAV’s
field we can see through our view). Let Rsensor be the FOV at t moment, then the motional tendency in the
max detection distance of UAV’s sensors onboard while period ½t, t + DtTEDM  can be predicted and located.
it is flying, and hence, FOV can be described as a circu- What’s more, if we regard the regions that obstacles
lar region (as shown in Figure 4) where the position of will (probably) reach as threat regions, a path without
UAV’s particle (xuav , yuav ) represents the center of circle any collision threat can be designed in ½t, t + DtTEDM 
and Rsensor means the radius of circular region only if we avoid the threat regions. This method is
applied through the path planning in urban space, thus
ðxFOV  xuav Þ2 + ðyFOV  yuav Þ2 4R2 sensor ð2Þ
a safe and flyable path can be generated to avoid
dynamic obstacles, and the TEDM is constructed by
sensing the obstacles in FOV as shown in Figure 6.
TEDM
We should avoid not only static but also dynamic
obstacles while planning path in such a limited environ- Path planning based on TEDM
ment as illustrated in Figure 1; thus, a new approach is
put forward using TEDM, aiming to fulfilling the
Reasons for the methods
demand of real-time path planning in harsh and special Traditional methods often regard obstacles are far from
environments. each other. They assume the obstacles are small and
In order to simplify the static obstacles, a horizontal slow. When encounter with irregular obstacles, it is
plane is chosen from stereoscopic space. As most of the hard to simplify them. Furthermore, the fast dynamic
buildings in urban space are vertical from bottom to obstacles require efficient algorithm. The A* algorithm
top and their shape and position in horizontal plane can guide the UAV, but it often has a collision with
are time-invariant, we can use two-dimensional space irregular obstacles. The breadth-first algorithm always
4 Measurement and Control

Thus, the local environment in FOV can be depicted

in TEDM, and the motion tendency of obstacles in
FOV can be predicated in a short time interval DtTEDM
     
xt + DtTEDM xt Vx
= + DtTEDM  ð6Þ
yt + DtTEDM yt Vy

Then, the TEDM in FOV is gridded and each grid

represents a minimum node when flying, so breadth-
first algorithm is used to explore the environment in
Figure 6. Time-environment dynamic map. FOV. Given kth local cost function gk as
X
n
gk = gki ðsi , ui Þ ð7Þ
searches the entire environment to find the optimal
i=m
path, causing a waste of calculating time.
We take the advantages of A* and breadth-first where m stands for the UAV’s node now, n means the
algorithms to solve the problems. The trace-back model number of nodes in FOV, si shows the position of UAV
is applied to find the optimal local path, purposing on and ui represents the decision-making when it comes to
reducing the time-consumption and improving the ith step.
safety. Finally, the method for path planning based on When the start node m is set, all flyable paths can be
TEDM is built, which can not only reflect to dynamic determined in FOV
obstacles fast but also avoid collision into irregular
obstacles. gk = gkm ðsm , um Þ + Gkm + 1, n ðsm + 1 , . . . , sn Þ ð8Þ

where Gkm + 1, n (sm + 1 , . . . , sn ) is the gathering that made

decisions from m + 1 stage to final n stage at kth path
Principium of path planning based on TEDM
planning, which can be denoted as Gkm + 1, n .
Generally speaking, the purpose of path planning is to The breadth-first algorithm is used to compute the
design a safe path for UAV; however, as we analyzed cost values of marginal nodes. When the decision-
in section ‘‘Introduction,’’ the existing methods can making process reaches FOV’s marginal nodes, the
hardly meet the requirements in such complex environ- optimal evaluation function can be selected by choos-
ments with crowded obstacles. As shown in Figure 7, ing the minimum fk of marginal nodes
an integrated approach can be planned as follows.      
First, we adopt the breadth-first algorithm to explore opt fk = opt gk + opt hk ð9Þ
the surrounding environment and then find out an
optimal waypoint in FOV with A* algorithm, and where opt is short for optimization, which means the
finally apply the trace-back model to plan an optimal minimum fk we choose.
path in each short time interval. In this way, we can get Take a step back from opt(gk ), so the former optimal
a path avoiding obstacle collisions over the urban step Gn1, n can be defined as
space. X
n1  
 
First, a global evaluation function is constructed by opt gk = gki ðsi , ui Þ + opt Gkn1, n ð10Þ
A* algorithm i=m

After equation (10), the optimal step Gkn1, n can be

fk = gk + hk ð4Þ
obtained, but the step between m + 1 and n  1 is not
hk = jend:x  now:xj + jend:y  now:yj ð5Þ the next step, we can infer opt gk equals to the
sure. InP
sum of in2 k k
= m gi (si , ui ) and opt Gn2, n . Then, the trace-
where (x, y) presents the position in two-dimensional back model can be applied by this analogy. And, an
space. end:x and end:y represents target position. now:x optimal path in FOV can be designed until it meets the
and now:y represents the current position of UAV. k is m node
the sequence number when building TEDM.

Figure 7. TEDM-based path planning system.

Zhao et al. 5

Table 1. Path planning step sequence.

Step The applied strategies

1 Build TEDM
2 Grid and label FOV
3 Search and record environment by
breadth-first algorithm
4 Select the optimal node using A* algorithm
5 Get path in FOV with trace-back model algorithm

TEDM: time-environment dynamic map; FOV: field of view.

  X
n1  
opt gk = gki ðsi , ui Þ + opt Gkn1, n
i=m
X
n2  
Figure 8. Obstacle sensing by sensors onboard.
= gki ðsi , ui Þ + opt Gkn2, n
ð11Þ
i=m
..
.
 
= gkm ðsm , um Þ + opt Gkm + 1, n

Finally, the above procedures are repeated until a

safe path in the global urban space is obtained. The dis-
tance of the final path can be expressed as
X
Dis UAV = Dis UAVk, k + 1 ð12Þ
k

Implementation of path planning

based on TEDM
As described in section ‘‘Path planning based on
TEDM,’’ the proposed method aims to plan an
obstacle-free path using the TEDM in urban space, Figure 9. Grid map of obstacles in UAV’s FOV.
and that can be achieved as following five steps
(Table 1).
As we discussed in section ‘‘TEDM,’’ all obstacles in
FOV can be predicated in ½t, t + DtTEDM  (see in
Figure 8) and then translated to TEDM. It is because
that DtTEDM is infinitesimal so that the UAV can avoid
all obstacles with avoiding the obstacles sensed in
½t, t + DtTEDM .
After translating the environment to TEDM, we
grid the environment. The A* algorithm, breadth-first
algorithm and trace-back model rely on grids to design
the path. Therefore, we transfer the environment to a
matrix at the beginning, and labeled 0 and 1 to the safe
area and obstacles, respectively. To TEDM, the labels
are as shown in Figure 9.
Then, breadth-first algorithm is used to explore and Figure 10. Explore and record environment based on breadth-
record all nodes in FOV from UAV’s start node, and first algorithm.
for every node at the margin of FOV, the cost function
is calculated to define the optimal margin node (see
Figure 10). Tracing back step by step, we can get the optimal path
After that, an optimal node on FOV’s edge can be in FOV at t moment.
obtained using A* algorithm, and take a step back from Finally, this integrated method is applied over the
this optimal node, then it is the second optimal node. environment along with the position of target and then
6 Measurement and Control

Figure 11. Functional architecture of TEDM-based path planning.

a safe flyable path can be designed to avoid collisions

in the urban space crowded with different kinds of
obstacles.
Figure 11 shows the functional architecture for path
planning based on TEDM and illustrates how this inte-
grated method operates.

Simulations and analysis

Suppose the complex environment in urban space is a
rectangular with 500 m height and 200 m width, the of
UAV velocity is 2 m/s, and its mission is to fly from the
start (2, 109) position, with passing through the first
target (76, 55) and the second target (450, 13), to the
final position (480, 101). Figure 12 shows the simulated
Figure 12. TEDM-based path planning.
results.
To further depict the safety of the UAV, some flying
moments are shown in Figure 13, where the brown disk

Figure 13. Some moments when designing path in urban space.

Zhao et al. 7

Table 2. Comparisons of A*, Voronoi and TEDM.

Methods Time Average memory Collision

consumption consumption (%) amounts
(min)

A* – 82 1
Voronoi – 91 14
TEDM 14.8 46 0

TEDM: time-environment dynamic map.

most important factor is safety; here, we use collision

Figure 14. Path planning based on A* algorithm compared amounts to judge the safety. We can infer from Table 2
with TEDM. that TEDM has no collision, while A* and Voronoi get
1 and 14, respectively. The results show that TEDM
has a better performance compared with the other two
traditional algorithms.

Conclusion
In terms of complex environment, we have argued how
the UAV can avoid collision in urban space safer and
faster. After reviewing the studies in the context of
immobile and dynamic obstacles, a conceptual con-
structive model which combines A* and breadth-first
algorithms has been proposed. Following are some
concluding remarks:

1. A TEDM is built, which divides the whole target

Figure 15. Path planning based on Voronoi compared with
into two parts, global target and local target. A*
TEDM.
and breadth-first algorithms are applied to glo-
bal and local environments, which improve the
safety in path planning.
stands for the FOV, and the gray blocks and the black
2. Small time interval is introduced to transfer
spots represent static and moving obstacles,
dynamic obstacles to instantaneous immobile
respectively.
obstacles, making it more efficient to avoid
As shown in Figure 13, a safe path can be obtained
dynamic obstacles.
with the method of TEDM. However, the effectiveness
3. The proposed method can avoid irregular build-
of TEDM is not discussed. So, we introduce A*- and
ings without simplification, using A* algorithm
Voronoi-based approaches to verify the effectiveness of
to guide itself globally and breadth-first method
the method of TEDM.
to avoid local obstacles.
As shown in Figure 14, the TEDM algorithm per-
forms well, but A* algorithm cannot arrive to the final
Still, there are several issues in need of attention and
position. Because A* algorithm tries to look for the
further investigations, including practical studies and
final position and attempts to get an optimal path from
the study on three-dimensional path planning.
the UAV to final position, and when there are threats
between the UAV and final position, it will probably
fall into dead zone. Declaration of conflicting interests
In Figure 15, though the Voronoi graph can plan a The author(s) declared no potential conflicts of interest with
path from start position to final position, but the path respect to the research, authorship, and/or publication of this
often goes through buildings. The TEDM algorithm article.
can avoid crowded obstacles safely, and the path
obtained by TEDM is smoother than Voronoi graph. Funding
More detailed information can be found in Table 2.
The author(s) disclosed receipt of the following financial sup-
The results show that TEDM costs 14.8 min in getting port for the research, authorship, and/or publication of this
a safe path, but the other two never arrive to the final article: The research was supported by the National Natural
position. To the average memory consumption, TEDM Science Foundation of China (grant nos 61503405 and
performs better than A* and Voronoi, which spends 61703428) and National Aeronautical Science Foundation of
less than half memory consumption. To the UAV, the China (grant nos 20160896007 and 20160896008).
8 Measurement and Control

ORCID iD Proceedings of the American control conference, Portland,

Xiaolin Zhao https://orcid.org/0000-0002-5094-9615 OR, 8–10 June 2000.
14. Fogel DB and Fogel LJ. Optimal routing of multiple
autonomous underwater vehicles through evolutionary
References programming. In: Proceedings of the 1990 symposium on
1. Amin JN, Boskovi JD and Mehra RK. A fast and effi- autonomous underwater vehicles technology, Washington,
cient approach to path planning for unmanned vehicles. DC, 5–6 June 1990, pp. 44–47.
In: Proceedings of the AIAA guidance, navigation and con- 15. Capozzi BJ and Vagners J. Evolving (semi)autonomous
trol conference, Keystone, CO, 21–24 August 2006. vehicles. In: Proceedings of the AIAA guidance, naviga-
2. McLain TW and Beard RW. Trajectory planning for tion, and control conference and exhibit, Montreal, QC, 6
coordinated rendezvous of unmanned air vehicles. In: August 2001.
Proceedings of the AIAA guidance, navigation, and control 16. Lai CK and Whidborne JF. Real-time trajectory genera-
conference and exhibit, Denver, CO, 14–17 August 2000. tion for collision avoidance with obstacle uncertainty. In:
3. Wang SL, Wei RX, Shen D, et al. Laguerre diagram con- Proceedings of the AIAA guidance, navigation and control
struction algorithm for path planning. Syst Eng Electron conference, Portland, OR, 8–11 August 2011.
2013; 35(3): 552–556. 17. Van den Berg J and Overmars M. Kinodynamic motion
4. Bortoff SA. Path planning for UAVs. In: Proceedings of planning on roadmaps in dynamic environments. In: Pro-
the American control conference, Chicago, IL, 28–30 June ceedings of the 2007 IEEE/RSJ international conference on
2000. intelligent robots and systems, San Diego, CA, 29 Octo-
5. Chandler PR, Rasmussen S and Pachter M. UAV coop- ber–2 November 2007.
erative path planning. In: Proceedings of the AIAA gui- 18. Zhang Q, Wei R, He R, et al. Path planning for
dance, navigation, and control conference and exhibit, unmanned aerial vehicle in urban space crowded with
Denver, CO, 14–17 August 2000. irregular obstacles. Control Theor Appl 2015; 32(10):
6. Wei RX, Xu ZF, Wang SL, et al. Self-optimization A- 1407–1413.
star algorithm for UAV path planning based on 19. Greatwood C and Richards A. Implementation of fast
Laguerre diagram. Syst Eng Electron 2015; 37(3): 577– MPS with quadrotor for obstacle avoidance. In: Proceed-
582. ings of the AIAA guidance, navigation and control confer-
7. Bellingham J, Richards A and How JP. Receding hori- ence, Boston, MA, 19 August 2013.
zon control of autonomous aerial vehicles. In: Proceed- 20. Techy L and Woolsey CA. Minimum-time path planning
ings of the American control conference, Anchorage, AK, for unmanned aerial vehicles in steady uniform winds. J
8–10 May 2002. Guid Control Dyn 2009; 32(6): 1736–1746.
8. Beard RW, McLain TW and Goodrich M. Coordinated 21. Singh SN, Pachter M, et al. Input–output invertibility
target assignment and intercept for unmanned air vehi- and sliding mode control for close formation flying. In:
cles. In: Proceedings of the IEEE international conference Proceedings of the AIAA guidance, navigation and control
on robotics and automation, Washington, DC, May 2002, conference, Denver, CO, 14 August 2000.
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.
1.6.6229&rep=rep1&type=pdf Appendix 1
9. Russell S and Norvig P. Artificial intelligence: a modern
approach. Upper Saddle River, NJ: Prentice Hall, 2003. Notation
10. Mitchell JS and Keirsey DM. Planning strategic paths
f, g global and local cost function
through variable terrain data. In: Proceedings of the SPIE
conference on applications of artificial intelligence, vol.
Gkm + 1, n decision-making gather from m + 1 stage
485, Arlington, VA, 14 June 1984, pp. 172–179. Belling- to final n stage at the kth path planning.
ham, WA: SPIE. Rsensor maximum detection distance of UAV’s
11. Stentz A. Optimal and efficient path planning for sensors onboard
!
partially-known environments. In: Proceedings of the V uav UAV’s velocity vector
1994 international conference on robotics and automation, x, y x and y positions of UAV
vol. 4, Los Alamitos, CA, 8–13 May 1994, pp. 3310–
3317. New York: IEEE. DtTEDM update rate of time-environment dynamic
12. Khatib O. Real-time obstacle avoidance for manipulators map
and mobile robots. Int J Robot Res 1986; 5(1): 90–98. u orientation of UAV
13. Vanek B, Péni T, Bokor J, et al. Practical approach to
real-time trajectory tracking of UAV formations. In: