INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 19, No. 9, pp.

1331-1338 SEPTEMBER 2018 / 1331

REGULAR PAPER DOI: 10.1007/s12541-018-0157-1

ISSN 2234-7593 (Print) / 2005-4602 (Online)

Level Set Based Path Planning Using a Novel Path

Optimization Algorithm for Robots

Xiao-Guang Zhang1, Wei Zhang1,#, Hui Li1,2, Ming-Qin Liu1, and Sungki Lyu3,#
1 School of Mechanical and Electrical Engineering, China University of Mining & Technology, 1, Daxue Road, Xuzhou, Jiangsu, 221116, China
2 School of Mechanical and Electrical Engineering, ZaoZhuang University, 1, Beian Road, Zaozhuang, Shandong, 277160, China
3 School of Mechanical and Aerospace Engineering, ReCAPT, Gyeongsang National University, 501, Jinju-daero, Jinju-si, Gyeongsangnam-do, 52828, Republic of Korea
# Corresponding Authors / E-mail: 18052127939@163.com, TEL +86-18052127939, ORCID: 0000-0002-5131-1621
E-mail: sklyu@gnu.ac.kr, TEL: +82-55-772-1632, ORCID: 0000-0001-9803-0272

KEYWORDS: Path optimization, Active contour, Path planning, Level set, Distance field

In order to decrease the path length and control the minimum distance between the path and the obstacles when the level set based
path planning algorithm is adopted , a new path optimization algorithm named elastic particle is proposed in this paper. Firstly, the
iteration expression of optimization algorithm is deduced by active contour theory. Secondly, to ensure the convergence of algorithm,
the relation among each item in the algorithm expression is analyzed and its convergence condition is determined. At last, level set
algorithm is improved so that the smoothness of the initial path and the convergence speed of the algorithm are improved. In addition,
a method named the nearest boundary distance is put forward to accelerate the operation speed of the algorithm. What’s more,
memory pool and binary sort tree are adopted in the code to further reduce the running time of this algorithm. The optimal values
of the algorithm parameters are analyzed via the simulation experiment,and its result demonstrates that the new algorithm has greatly
optimized the path of algorithm-level set and guaranteed fast running speed and high reliability.

Manuscript received: July 28, 2017 / Revised: March 19, 2018 / Accepted: June 15, 2018

1. Introduction to solve continuous environments prone to currents issues. Mitchell et

16
al. studied the path planning problems which satisfy several additive
17
Path planning plays an important role in robot navigation. Generally, constrains. Kimmel et al. proposed a new FMM algorithm to solve
the workspace of the robot consists of free space and obstacle space. the 3D reconstruction problem and the new FMM algorithm combined
The essence of path planning is searching a path in the free space when with a back tracking gradient descent along the reconstructed surface to
1
meeting certain constraints. Common methods of path planning include search the path in robot navigation.
2-5 6-8
particle swarm optimization (PSO), genetic algorithm (GA), According to cost distribution computed by the 1st step, common
9,10
simulated annealing algorithm (SAA) and ant colony algorithm method used to search the path is gradient descent algorithm. The
11-13
(ACA). shortest path search method has been applied more and more widely in
18-20
In recent years, many researchers have focused on the path planning path planning field in recent years, such as A* algorithm, Dijkstra
21,22
algorithm based on level set because of its rapid running speed and algorithm.
implementation simplicity. It can be divided into two main parts: (1) Though the path planning algorithm based on level set is fast in
obtaining the global cost distribution of the target point via solving calculation and more suitable for robot navigation with high real-time
Eikonal equation; (2) getting the path via greedy method according to requirement, there are some disadvantages in this algorithm:
the cost distribution. (1) The path obtained via the level set method is not optimal. Its
For the first step, most researchers have proposed a variety of length and smoothness need to be further improved.
14
improved solutions to accelerate the solving process. Xu et al. formally (2) The shortest distance between the path and the obstacle couldn’t
analyzed the level set algorithm and researched the mechanisms of the be adjusted and thus it cannot fit robots of various shapes and
15
value of the level set function changing with environment. Petres et al. dimensions.
proposed a novel Fast Marching Method based path planning algorithm Therefore, an elastic particle path optimization algorithm is proposed

© KSPE and Springer 2018

 1332 / SEPTEMBER 2018 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 19, No. 9

to further optimize the initial path by analyzing the active contour theory.
The basic principle of this algorithm is that the path is considered as a
set of ordered particles. Each particle is balanced under the action of
internal and external force so that the path converges to a predetermined
position. In the construction of external force, in order to accelerate the
running speed of this algorithm, memory pool and binary sort tree are
used for the implementation of the resolving distance field problem. To
accelerate the convergence speed of optimizing algorithm, bilinear
interpolation and four-order Range-Kutte method are taken into the
velocity express in Eikonal function to solve the initial path.

Fig. 1 An example of distribution distances between points in free

2. Path Optimization Algorithm space and obstacle space

2.1 Theoretical deduction of the algorithm

The path obtained by the previous method can be processed further 2.2 Convergence condition of the optimization algorithm
to decrease the length and increase the smoothness. A path optimization When ∂C/∂t = 0 , the change rate of the path is zero, which means
algorithm inspired from the active contour is proposed. The basic idea the path optimization algorithm convergent to the desired position. Eq.
of this algorithm is that the path is considered to consist of several (3) can be modified as follows:
particles, the position of every particle is determined by internal force
and external force, so we name this algorithm elastic particles algorithm. – (β C′)′ + ∇g = 0 (6)
Suppose a curve C(p), the aim is to obtain the shortest path. So this
problem can be transformed into following energy functional problem: Following conclusion can be achieved from Eq. (6), the resultant
force of each particle is zero and all particles stop motion, then the path
1 1 1
2 2
E ( C ( p) ) = ∫ ω 1 Cp ( p) dp + ∫ ω 2 Cpp ( p) dp + ∫ g ( C ( p ) ) dp (1) will converge to the desired state. In order to ensure the convergence
0 0 0
of our algorithm, the most important premise is making appropriate
1 1
The second item ∫ Cpp ( p) dp = ∫ κ ⎛ -------⎞ dp is the oscillation
2 2 dC
choice of selecting the appropriate external force.
0 0 ⎝ ds ⎠
energy of the curve, the first item is the length of the curve. According Eq. (7) is new repulse potential field.
to curvature motion evolution theory, the curve will become smoother
⎧ ⎛ d ( x, y )⎞
by decreasing the length of curve, and accordingly the second can be ⎪δ 1 – --------------
- d ≤ Rth
P ( d ( x, y ) ) = ⎨ ⎝ Rth ⎠ (7)
eliminated. Eq. (1) can be simplified as follows: ⎪0
⎩ d > Rth
1 1
2
E ( C ( p)) = ∫ β Cp ( p ) dp + ∫ g ( C ( p) ) dp (2) where d(x, y) represents the distance from point (x, y) in the free space
0 0
to the forbidden space.
The gradient descent flow of Eq. (2) is Thus, the repulse force can be set according to Eq. (8)

∂C ∇P
------- = – (β C′)′ + ∇g (3) ∇g = Frep = –δ ----------- (8)
∂t ∇P

The initial path can be optimized by Eq. (3). The first item is the In which δ is the magnitude of the repulse force. ∇P/ ∇P is the
internal force which can be classified as interaction force between normalized potential field, which is the direction of the repulse force.
particles, and the second item is the external force exerted by the Hence, in order to guarantee the converge of algorithm, the path
external environment on the particles. expression must meet the following equation:
Eq. (3) is discretized through finite differences method and the
δ
recursion formulae of path can be got C″ ≤ –--- Frep ( d ) (9)
β
n+1 n n n n n n
Ci = Ci – Δt (β ( ( Ci – Ci – 1) – ( Ci + 1 – Ci ) ) + ∇g ( Ci ) ) (4) Eq. (9) can also be rewritten as follows

δ ⎧ ∇P ⎫ δ
Where i represents the ith particle, n is the iterative number, C is C″ ≤ --- sup ⎨ ----------- ⎬ = --- (10)
0
β Ω ⎩ ∇P ⎭ β
the initial path which is obtained by the previous algorithm. C consists 0

n n
of m particles, Co is the start point, Cm is the destination. Therefore, Eq. (10) can guarantee the convergence of our algorithm.
Let g(C) = −P(C), the P(C) can be seen as the repulse potential in
the artificial potential algorithm, ∇P ( Ci ) is the repulse force acting on
the ith particle. So the equation can be modified as: 3. The Solving Method of Distance Field

n+1 n n n n n
Ci = Ci – Δt (β ( ( Ci – Ci – 1 ) – ( Ci + 1 – Ci ) ) – Frep ) (5) In section 2, the distance from the point in free space to the obstacle
 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 19, No. 9 SEPTEMBER 2018 / 1333

space is represented by d(x, y). The d(x, y) can be calculated by

following analysis.
In Fig. 1, the distribution of d(x, y) can be described by distance
field. Thus d(x, y) can be expressed by the level set formation as shown
in Eq. (11):

Γ ( c ) = {{x, y }|d ( x, y )=c, ( x, y ) ∈ Ωfree } (11)

Γ(c) represents the distance field. There are two main methods
currently for getting the solution of distance field set: the direct method
and the Eikonal equation method.
(1) The direct method is easy to operate and realize, which is
described in Eq. (12):
k k k k
di = min {di1, di2, …, dM }
(12)
1 2 N
di = min {di , di , …, dij }
Where i is the ith grid point in free space, j is the jth boundary point, Fig. 2 The initial status of computing distance field set
k is the kth obstacle. The complexity of this method is O(MN), it requires
a great number of computations.
(2) Eikonal equation method gets the distance distribution through
solving Eikonal equation shown as Eq. (13):

1
∇T = --- (13)
V

Set V=1, according to the definition of Eikonal equation, the initial

value is the set of obstacle boundary point, the solution of Eikonal
equation is the distance field set. However, the distance obtained Fig. 3 The updating method when there exists rational distance
through this method is not really Euclidean distance. Thus, a method
named nearest boundary point diffusion based on Fast Marching
Method (FMM) is proposed in this paper to establish distance field set.
Similar to FMM, this method divides the network points into three
types: frozen, narrow band and far, and adds a nearest source domain
for each point to store the nearest boundary point. Steps are as follows:
(1) Transforming the input map into a binary image of double types.
Wherein, the obstacle space value is 1 and the free space value is 0.
The map boundary point is extracted through Eq. (14):
Fig. 4 One updating method when there exists two rational distance
map ( i, j ) = 1&& ( map ( i + 1, j ) = 0 ||map ( i – 1, j ) = 0 values
(14)
||map ( i, j + 1 ) = 0 ||map ( i, j – 1 ) = 0)

(2) The blocks are emblems of pixels, in which of numbers are

symbol of distance to obstacles. White blocks present obstacle whose free space to initial point of arrowhead chain, is the length between free
boundary is showed by the light grey, and free space is indicated by space point and obstacle. In other wards, it is a minimum distance for
dark grey, as showed by Fig. 2(a). The boundary point is marked as point in free space to any boundary point.
frozen point and the distance is set to 0, and the 0 point is showed by (3) Selecting the point with shortest distance in the set of all narrow
Fig. 2(a) Other points in free space belong to far set and the distances band points and set it as frozen point. Then the distances of its four
are infinite, as shown in Fig. 2(a) by the ∞ point. We search four neighboring points are updated as below:
neighboring points of every boundary point. If the neighboring point is A. As shown in Fig. 3, if q is the point with shortest distance
a free space point and belongs to far point set, this neighboring point selected from the narrow band set, the distance of p needs to be
will be set as a narrow band point and its distance value will be set to updated. In this case, there is rational distance (rational distance means
1. The blocks as a range of 1 to symbol search for every boundary the distance value is not equal to ∞) value only in one direction of point
point, which is the nearest boundary point of its neighboring point, as p, thus its distance value will be updated as dq+1. The nearest boundary
shown in Fig. 2(b). From the Fig. 2(b), the boundary point a is nearest point of p is same as the nearest boundary point of q.
boundary point of its neighboring point b and c. Arrowhead chain B. Suppose that there are rational distance values in two directions,
directing transmition direction is formed by some steps, which initial assume (xC, yC) is the point whose distance needs to be calculated. Set
stage presents the boundary of obstacle. The distance, from point in a = min ( dxC – 1, yc, dxC + 1, yc ) , b = min ( dxC, yC – 1, dxC, yC + 1 ) and c =
 1334 / SEPTEMBER 2018 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 19, No. 9

Fig. 5 Another updating method when there exists two rational Fig. 6 Simulation map (For estimating the influence of β and Δt on
distance values the convergence of path optimization algorithm)

min(a, b). The distance value of this point will be updated by the evolution stage. External force does not work, the path evolution is
following methods: completely controlled by internal force according to Eq. (16)
As shown in Fig. 4, q is the point with shortest distance selected 2
∂C ∂ C
from the narrow band set, the distance of p is to be updated. If a==b, ------- = --------2- (16)
∂t ∂p
S is the nearest boundary point of the point with rational value in
1

direction x, S is the nearest boundary point of the point with rational

2 The energy functional will be
value in direction y, d is the distance between S1 and p, d is the
1 2
1 D
2
distance between S and p, then the distance value of point p will be
2 E ( C ( p )) = ∫ Cp ( p ) dp = ∫ ds (17)
0 S
updated as min(d , d ). The nearest boundary point of point p is Sseq,
1 2

seq = argmin(d , d ).
1 2 Eq. (17) is gradient descent flow of Eq. (16). Eq. (17) illustrates that
As shown in Fig. 5, q is the point with shortest distance selected decreasing the energy means the decrease of length of C(p). The path
from the narrow band set, the distance of p is to be updated. If a ≠ b , will evolve to a straight line.
S is the nearest boundary point of the point with rational value in
1 When there is a particle ei, which satisfies d(ei) ≤ rth, the path
direction x, S is the nearest boundary point of the point with rational
2 optimization is in the constrained evolution stage. In this stage, the
value in direction y. Set seq = argmin(a, b), so the nearest boundary position of particle will be gradually stable. Therefore, just selecting
point of point p is Sseq and the distance of point p is ||p−Sseq||. appropriate β and δ can ensure that the path will converge to the desired
position. In this paper, we propose the following method to determine
β and δ.
4. The Acquisition of Obtaining Initial Path
δ
--- = sup { C″ }n – 1 (18)
β n Ω
Suppose there is a velocity gradient area whose radius is Rth (Rth >
rth) around the forbidden space. In the algorithm of this paper, the map Where n is current iteration.
is a binary double-type picture, and the area with value 0 refers to free
area and the area with value 1 refers to forbidden area. Below velocity
function is proposed according to these information: 6. Experiment and Discussion
max ( Rth – d ( x, y )0 )
V ( x, y ) = 1 – -------------------------------------------
- (15) 6.1 Influence of Δt and β on convergence performance of path
Rth
optimization algorithm
Where d(x, y) is the distance of the point from free space to In Eq. (5), Δt and β are two important parameters. It is intuitive that
forbidden space. Eq. (15) is used to compute the solution of the Eikonal the convergence capability of path optimization algorithm is related to
equation. In the new method, the minimum distance between the path these two parameters. We will discuss how Δt and β impact on the
and the forbidden space is Rth. convergence performance of path optimization algorithm in this part.
Fig. 6 is the simulation map. Main parameters are set as follows: R = th

5, r = 3. The simulation scenes are shown in Table 1.

th

5. Convergence Analysis of the Optimization Algorithm Figs. 7 and 9 show the change of path length when β = 1 and Δt is
varying. Figs. 8 and 10 show the change of path length when Δt = 0.1
By the previous analysis, the path optimization consists two stages: and β is varying. Simulation experiments show that increasing Δt and
free evolution stage and constrained evolution stage. Suppose C is the β is conducive to decreasing the path length and accelerating the
initial path, e is a particle in the path, d represent the distance field, and convergence speed. There are two main reasons: (1) The increment of
n is the number of particles. Δt and β causes larger internal force, accordingly, the convergence
When rth < d(ei) ≤ Rth, i = 1, 2, …, n, initial path will be in the free speed is increased. (2) The larger Δt and β is, the longer movement
 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 19, No. 9 SEPTEMBER 2018 / 1335

Fig. 7 The change of length of path using Fig. 6(a) (Varying Δt, while Fig. 9 The change of length of path using Fig. 6(b) (Varying Δt, while
β = 1) β = 1)

Fig. 8 The change of length of path using Fig. 6(a) (Varying β, while Fig. 10 The change of length of path using Fig. 6(b) (Varying β,
Δt = 0.1) while Δt = 0.1)

Table 1 Simulation parameters (For estimating the influence of β and

Δt on the convergence of path optimization algorithm)
Start Destination
Fig. 6(a) (102,177) (278,415)
Fig. 6(b) (43,49) (283,237)

distance of particle will be, which will cause the smoothness and length
of the path decreases. However, when the value exceeds permitted
value, the feasible path may not be obtained. So we set β = 1, Δt = 0.1 Fig. 11 The distance distribution between every point on the path and
in this paper. the forbidden region when adopting Fig. 19(a)

6.2 Analysis of the convergence condition of path optimization

algorithm
The convergence condition of path optimization is analyzed in the
previous section. In this section, the effectiveness will be discussed.
The simulation map is presented in Fig. 6(a). Main parameters are
listed as follows: β = 1, Δt = 0.01. Iteration number is 10000. δ is
determined by Eq. (18).
The value of rth is 3, which means that the desired minimal distance
between the path and forbidden space is 3. From Fig. 11, the minimal
distance between the path and forbidden space is 2.9577. Fig. 12 Fig. 12 The change of max internal force when adopting Fig. 19(a)
illustrates that evolution of the path fits the previous analysis. Fig. 13
shows the initial path indicated by blue and final path indicated by red.
It can be got from these figures that the path optimization can make the the real-time capability of our algorithm is evaluated by comparing
initial path converge to the desired position. with the Eikonal equation solution method proposed in Ref. 23. The
scene of this evaluation is shown in Fig. 14. From the figure, for one
6.3 Running time of path optimization algorithm via distance obstacle, maintaining the dimension of middle white obstacle and
field changing the size of black free area changes the measurement of
For the sake of speeding up the running rate of establishing distance simulation circumstance. The capability of algorithm to map can be
field set, memory pool and binary sort tree are adopted. In this section, checked on the constant dimension and changeable free area. The size
 1336 / SEPTEMBER 2018 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 19, No. 9

Fig. 15 Simulation scene for evaluating the time of solving distance

field

Fig. 13 The path obtained by our algorithm when adopting Fig. 19(a)

Fig. 16 Comparison of the time used for several distance field solving
Fig. 14 The scene of simulation map algorithms

of obstacle remains unchanged, while the size of free space is modified. 7. Conclusions
Under different size of maps, algorithm runs 50 times and record the
average running time. Fig. 15 is the parameters of this experiment, the In this paper, a new path optimization algorithm is proposed. Main
results are shown as Fig. 16. It can be seen that the memory pool and conclusions draw by previous analysis are presented as follows:
binary sort tree improve the speed of the stablishing distance field set (1) Compared with traditional Eikonal method, the nearest boundary
algorithm. The results meet the design goal of the algorithm. point propagation method can obtain the distance field distribution with
the minimum amount of time, which mainly owing to the adoption of
6.4 Running time of the whole algorithm memory pool and binary sort tree in this algorithm.
The running time of establishing distance field set is discussed in (2) A gradual change speed for Eikonal can reduce the oscillation
the previous section. In this section the running time of the whole phenomenon of initial path obtained by common gradient descent and
algorithm will be investigated. Fig. 17 is the simulation parameters. accelerate the convergence speed of the optimization algorithm. From
Other parameters are: β = 1, Δt = 0.1, R = 5, r = 3. The iteration
th th the simulation results, the error between minimum distance to obstacle
number is 5000. Algorithm runs 100 times under different scene and for every point and the expected is 1.41%, and it basically meets the
records the average running time. The results are shown as Fig. 18. The requirements.
establishment of the distance field set and the establishment of the (3) The convergence condition of the optimization algorithm can
global cost are only related to the map, which is independent of the ensure the convergence performance of the path. The final path
path in the scene. The simulation shows that the algorithm in this paper converges to the desired position under the influence of this
has high real-time performance. optimization algorithm. The simulation results show that the proposed
 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 19, No. 9 SEPTEMBER 2018 / 1337

Based Path Planning,” Proc. of the International Workshop on

Planning Under Uncertainty for Autonomous Systems, International
Conference on Automated Planning and Scheduling (ICAPS), pp. 9-
18, 2005.

2. Qin, Y.-Q., Sun, D.-B., Li, N., and Cen, Y.-G., “Path Planning for
Mobile Robot Using the Particle Swarm Optimization with Mutation
Operator,” Proc. of International Conference on Machine Learning
and Cybernetics, pp. 2473-2478, 2004.

3. Roberge, V., Tarbouchi, M., and Labonté, G., “Comparison of Parallel

Genetic Algorithm and Particle Swarm Optimization for Real-Time
UAV Path Planning,” IEEE Transactions on Industrial Informatics,
Vol. 9, No. 1, pp. 132-141, 2013.

4. Fu, Y., Ding, M., and Zhou, C., “Phase Angle-Encoded and
Quantum-Behaved Particle Swarm Optimization Applied to Three-
Dimensional Route Planning for UAV,” IEEE Transactions on
Systems, Man and Cybernetics, Part A: Systems and Humans, Vol.
42, No. 2, pp. 511-526, 2012.

Fig. 17 Start point and destination of the scene for evaluating running 5. Xue, Y. and Liu, H., “Optimal Path Planning in Complex Indoor
time Environment Based on Improved PSO,” Journal of Computational
Information Systems, Vol. 7, No. 6, pp. 2158-2165, 2011.

6. Ergezer, H. and Leblebicioglu, K., “Path Planning for UAVs for

Maximum Information Collection,” IEEE Transactions on Aerospace
and Electronic Systems, Vol. 49, No. 1, pp. 502-520, 2013.

7. Mohanta, J. C., Parhi, D. R., and Patel, S. K., “Path Planning

Strategy for Autonomous Mobile Robot Navigation Using Petri-GA
Optimisation,” Computers & Electrical Engineering, Vol. 37, No. 6,
pp. 1058-1070, 2011.

8. Cheng, C.-T., Fallahi, K., Leung, H., and Chi, K. T., “A Genetic
Algorithm-Inspired UUV Path Planner Based on Dynamic
Programming,” IEEE Transactions on Systems, Man, and
Cybernetics, Part C (Applications and Reviews), Vol. 42, No. 6, pp.
1128-1134, 2012.

Fig. 18 The results of operating time 9. Wang, X. C., Zhou, M. G., Liu, X. Y., and Fan, Y. C., “The Research
on Robot Global Path Planning Based on Simulated Annealing
Dissipative Ant System,” Proc. of 7th World Congress on Intelligent
algorithm can reduce the average time by 59.9% compared with the Control and Automation, pp. 5764-5769, 2008.
traditional algorithm.
10. Miao, H. and Tian, Y.-C., “Robot Path Planning in Dynamic
Environments Using a Simulated Annealing Based Approach,” Proc.
of International Conference on Control, Automation, Robotics and
ACKNOWLEDGEMENT
Vision, 2008.

The authors are grateful to the anonymous referees for their valuable 11. Duan, H., Yu, Y., Zhang, X., and Shao, S., “Three-Dimension Path
comments and suggestions for improving the presentation of this paper. Planning for UCAV Using Hybrid Meta-Heuristic ACO-DE
Project Funded by the Priority Academic Program Development of Algorithm,” Simulation Modelling Practice and Theory, Vol. 18, No.
Jiangsu Higher Education Institutions (PAPD). Production, Prospective 8, pp. 1104-1115, 2010.
Joint Research Projects in Jiangsu Province (BY2016026-02).
12. Zhang, G.-l. and Jia, H.-m., “Global Path Planning of AUV Based
on Improved Ant Colony Optimization Algorithm,” Proc. of IEEE
International Conference on Automation and Logistics (ICAL), pp.
REFERENCES
606-610, 2012.

1. Ferguson, D., Likhachev, M., and Stentz, A., “A Guide to Heuristic- 13. Wei, Z., Luo, Q., and Su, X., “A Novel Mobile Robot Path Planning
 1338 / SEPTEMBER 2018 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 19, No. 9

Method Based on ACO Algorithm Using New Ants Meeting

Judgment Strategy,” JDCTA: International Journal of Digital Content
Technology and its Applications, Vol. 6, No. 1, pp. 302-309, 2012.

14. Xu, B., Stilwell, D. J., and Kurdila, A. J., “Fast Path Re-Planning
Based on Fast Marching and Level Sets,” Journal of Intelligent &
Robotic Systems, Vol. 71, Nos. 3-4, pp. 303-317, 2013.

15. Petres, C., Pailhas, Y., Patron, P., Petillot, Y., Evans, J., and Lane, D.,
“Path Planning for Autonomous Underwater Vehicles,” IEEE
Transactions on Robotics, Vol. 23, No. 2, pp. 331-341, 2007.

16. Mitchell, I. M. and Sastry, S., “Continuous Path Planning with

Multiple Constraints,” Proc. of 42nd IEEE Conference on Decision
and Control, pp. 5502-5507, 2003.

17. Kimmel, R. and Sethian, J. A., “Optimal Algorithm for Shape from
Shading and Path Planning,” Journal of Mathematical Imaging and
Vision, Vol. 14, No. 3, pp. 237-244, 2001.

18. Pal, A., Tiwari, R., and Shukla, A., “Modified A* Algorithm for
Mobile Robot Path Planning,” in: Soft Computing Techniques in
Vision Science, Patnaik, S., Yang, Y. M., (Eds.), Springer, pp. 183-
193, 2012.

19. Krawiec, B., Kochersberger, K., and Conner, D. C., “Autonomous

Aerial Radio Repeating Using an A*-Based Path Planning Approach,”
Journal of Intelligent & Robotic Systems, Vol. 74, Nos. 3-4, pp. 769-
789, 2014.

20. Witt, J. and Dunbabin, M., “Go with the Flow: Optimal AUV Path
Planning in Coastal Environments,” Proc. of Australian Conference
on Robotics and Automation, 2008.

21. Liu, H., Stoll, N., Junginger, S., and Thurow, K., “A Floyd-Dijkstra
Hybrid Application for Mobile Robot Path Planning in Life Science
Automation,” Proc. of IEEE International Conference on Automation
Science and Engineering (CASE), pp. 279-284, 2012.

22. Shaikh, E. A. and Dhale, A., “AGV Path Planning and Obstacle
Avoidance Using Dijkstra's Algorithm,” International Journal of
Application or Innovation in Engineering & Management (IJAIEM),
Vol. 2, No. 6, pp. 77-83, 2013.

23. Hassouna, M. S. and Farag, A. A., “Multistencils Fast Marching

Methods: A Highly Accurate Solution to the Eikonal Equation on
Cartesian Domains,” IEEE Transactions on Pattern Analysis and
Machine Intelligence, Vol. 29, No. 9, pp. 1563-1574, 2007.