2019 IEEE 3rd Advanced Information Management,Communicates,Electronic and Automation Control Conference (IMCEC 2019)

AGV Path Planning based on Improved A-star

Algorithm
Tao Zheng1, Yanqiang Xu1,Da Zheng2
1. College of Electrical and Automation Engineering, Hefei University of Technology,
2. An’Hui HangDa Intelligent Technology Co. Ltd
Hefei, China
zhengtao@ustc.edu, 473421737@qq.com, 1411539832@qq.com

Abstract—With the wide application of Automated Guided The A-star algorithm only considers the actual value and
Vehicle (Hereinafter referred to as AGV) in modern the heuristic value of the node n so that there may be multiple
manufacturing, logistics and transportation industries, nodes n with the smallest F(n) value, and it takes no account
improving the work efficiency of AGV has become a major focus for the number of inflection points in the path planning which
of the industry research. In order to reduce the limitation of two may lead to many inflection points in the search path.
factors—AGV optimal running path search and path search Therefore, there is no guarantee that one can find the optimal
speed on the efficiency of AGV, this paper based on the A-star path using A-star algorithm. In this paper, the angle evaluation
algorithm uses the characteristics of jump point search to
cost function is added to the cost function of A-star algorithm
improve the node search mode and the search speed, and adds
to find the path with the least inflection point, and the jump
the angle evaluation cost function to the cost function of A-star
algorithm to find the path with the least inflection point, so as to
point search is used to reduce the number of node traversals in
quickly find the optimal path. And the simulation experiments in the open list by the A-star algorithm to speed up the path
this paper show that the optimized A-star algorithm can search search in the grid map.
for the optimal path, of which path search speed is faster than
that of the A-star algorithm. II. A-STAR ALGORITHM IMPROVEMENT

Keywords—AGV; A-star algorithm; jump point search; path A. A-STAR Algorithm Cost Function Improvement
planning;heuristic function
In the factory environment which usually has regular layout
of north-south and east-west, the Manhattan distance is closer
I. INTRODUCTION to the actual distance, and makes the algorithm more efficient.
Since 2014, AGV research has ushered in a breakthrough So in the improved A-star algorithm, Manhattan distance is
point [1]. AGV continues to innovate and is widely used in chosen as the heuristic function, then H1(n) is:
digital factories, warehouse handling systems, and flexible
manufacturing systems. Improving the efficiency of AGV has H1 (n) =| xn − xg | + | yn − yg | (1)
always been the focus of research. Optimizing AGV path The biggest problem with the A-star algorithm cost
planning is the preferred direction to improve the efficiency of function search is that it cannot select the optimal path with the
AGV. The path with the least and shortest inflection point in least inflection point in many symmetric paths with the same
AGV path planning is called the optimal path. The optimal path length [12]. In this paper, the A-star algorithm is improved
path can not only improve the transportation efficiency of the by combining the angle cost estimation method and the node
AGV in the working process, but also reduce the damage of the culling method based on moving omnidirectionality. Assuming
AGV itself, which is of great significance for improving the
that the node n’s coordinate is（xn, yn ）, then the previous
overall production efficiency and reducing the production cost.
node’s coordinate is（xn-1, yn-1 ）,and the coordinates of the
There are some common path planning algorithms such as next node is（xn+1, yn+1）. Then the angle evaluation function
Dijkstra algorithm [2], best priority search [3], A-star algorithm of node n is H2(n)：
[4], D-star algorithm [5], and intelligent algorithms like genetic
algorithm [6], ant colony algorithm [7], particle swarm H 2 ( n) =
( xn −1 − xn )( xn − xn +1 )+( yn −1 − yn )( yn − yn +1 ) (2)
algorithm [8], etc. Among so many path planning algorithms, ( xn −1 − xn )2 + ( yn −1 − yn ) 2 + ( xn − xn +1 )2 +( yn − yn +1 ) 2
A-star algorithm is the most widely used for approaching a Combining the heuristic function and the angle evaluation
near optimal solution, fast solution speed and high efficiency, function, the search function of improved A-star algorithm is as
which still needs to be improved. Wei Shan et al.[9] improved follows:
the A-star algorithm heuristic function to design a search
algorithm that adapts to the factory path but did not consider F(n)= G(n)+ k1 H1(n)+ k2 H 2 (n) (3)
the path inflection point and search speed. Li Weiguang [10] In the above function, n is the next node of the current node,
improved the JPS algorithm to find the path with the least while F(n) represents the estimated value from the initial node
inflection point, but did not take the length of the path into passing through the node n to the target point, and G(n)
account. represents the estimated value from the initial node to the node

978-1-7281-0513-0/19/$31.00 ©2019 IEEE 1534

Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 14,2023 at 12:02:46 UTC from IEEE Xplore. Restrictions apply.
 n. Since their units are different, H1(n) and H2(n) need to be C. Improve A-star Algorithm Flow
normalized by normalization parameters. Here k1 and k2 refer Start
to the normalized parameter of the function.
The improved A-star algorithm, which added the angle Create Open_List and
Close_List
evaluation cost function, prioritizes the path with the least
inflection point in the search process, so that an optimal path
Add the starting point P0
suitable for AGV motion can be found. to the Open_List

B. A-star Algorithm Search Preprocessing Is the current node Pi a

Add the current
node Pi and the

The jump point search is to find the point that can jump
target node Pn? Yes target node Pn to
the Clost List

between the starting point and the target point, and find the No

path by searching between jump points. This is the search

Add P0 to the Clost List, and
the adjacent node Pi with the

strategy of the jump point search algorithm. As shown in Fig. 1, smallest F value is added to
the Open List.

the G point can be directly searched from the S point, without The optimal node Pj
passing through the nodes n2 or n4. The intermediate nodes on
Jump search for the jump
is taken as the point of the current node
current jump node Pi
the path formed by the jump points (like S and G ) are not Pi direction

expanded, which also greatly reduces the amount of calculation

and storage during the expansion process and speeds up the Whether there is a
jumpable node No
There is no feasible
path between Pi

searching. Yes
and Pn

Add the current node to

Add the optimal
the Clost List And add the
node Pj to the Clost
jumpable node to the
List
n1 n2 n6 Open List

Is there a target node Add the target node

Pn in the Open List? Yes Pn to the Clost List

S n3 G No

Traverse the nodes in the

Open List and find the F-
value minimum node Pj End

n5 n4 n7 Fig.2 Improved A-star algorithm search flow chart

III. SIMULATION
Fig.1 Extension process of nodes during jump point searching
In order to prove the advantages of the improved A-star
For the conventional A-star algorithm, it takes a lot of time algorithm in search speed and path selection, two different
to evaluate the node value. Here the jump point search strategy experimental environments are set: 20*20 blank gird map,
is used to preprocess the nodes in the open list of the A-star 20*20 set gird map. The search performance of two
algorithm, which reduces the evaluation process of unnecessary algorithms are compared by comparing searching path in the
nodes and achieves the purpose of speeding up the search. maps. From many experiments, it is obtained that the cost
function parameters are k1=2, k2=2.
Preprocessing rule: Use the cost function to find the node
n1 with the smallest F(n) around the initial point as the next A. Blank Grid Map Search
node of the path planning, and search along the initial node to The experiment establishes a 20*20 blank map to simulate
the node n1 (called the priority direction principle), and use it the AGV working environment, and sets a starting point
in this process. The principle of jump point, find the jump point coordinate and a target point coordinate. The search ways and
in this direction, and join the open list to add the last jump the result comparison of the two algorithms (A-star algorithm
point to the closed list. With the new jump point as the current and improved A-star algorithm ) in this environment are shown
node, the search is repeated until the target point is found, the in Fig.3 and TABLE I.
search ends and the path is displayed. Or the target point was
not found and the open list is empty and the search failed.

(a) A-star algorithm path search (b) Improved A-star algorithm path search
Fig.3 Path search for 20*20 blank map environment

1535

Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 14,2023 at 12:02:46 UTC from IEEE Xplore. Restrictions apply.
 TABLE I. COMPARISON OF PERFORMANCE OF TWO ALGORITHMS

Inflection
Algorithm Start point End point Path length Search time
point
A-star （7，8） （15，16） 16 3.732s 6
Improve A-star （7，8） （15，16） 16 0.434s 1
It can be seen from Fig.3 that both the A-star algorithm
and the improved A-star algorithm can search for symmetric
paths of the same length, but the search path of the A-star
algorithm has more inflection points, and it is obvious that the
improved A-star algorithm is superior to the search path. At
the same time, the time to search for the target point by both
shows that the improved A-star algorithm search faster than
the A-star algorithm by more than 5 times.
B. Set Gird Map Search
a b
In the experiment, a gird map with obstacles is established
simulating the working environment of the AGV, in which
there is a starting point and a target point and the black square
means no thoroughfare. The A-star algorithm and the improved
A-star algorithm are used to search path in the gird map.
a) Comparison of Path Planning for the Same Starting and
Ending Point
This experiment sets a fixed starting and target point and
uses the A-star algorithm to plan path multiple times, and
records the experimental data, as shown in Fig.4 below:
c d
Fig.4 A-star algorithm path planning

TABLE II. A-STAR ALGORITHM SEARCH PERFORMANCE

Experiment Start point End point Path length Search time Inflection point
a （5，5） （20，20） 30 3.714s 3
b （5，5） （20，20） 30 4.373s 4
c （5，5） （20，20） 30 4.417s 5
d （5，5） （20，20） 30 4.104s 3
As shown in Fig.4, the paths obtained by multiple
experiments with the same starting point and target point have
the same length, but the path trajectory is totally different, and
the number of inflection points in the paths are also different.
It is difficult for the A-star algorithm to search for optimal
path.
In the same search environment, the improved A-star
algorithm is used to search path four times, of which the
search trajectory and the search time are recorded in the
following TABLE III.
Fig.5 Improved A-star algorithm path planning

TABLE III. IMPROVE A-STAR ALGORITHM SEARCH PERFORMANCE

Experiment Start point End point Path length Search time Inflection point
a （5，5） （20，20） 30 0.539s 1
b （5，5） （20，20） 30 0.569s 1
c （5，5） （20，20） 30 0.520s 1
d （5，5） （20，20） 30 0.555s 1

1536

Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 14,2023 at 12:02:46 UTC from IEEE Xplore. Restrictions apply.
 According to four times experiment data, the obtained path
trajectories are the same with the least inflection point, which
indicates that the improved A-star algorithm has stable search
performance and can search for the optimal path. And
comparing the search time in TABLE II and III, the improved
A-star algorithm search faster than the A-star algorithm by 5
times under the same condition.
b) Comparison of Path Planning for Different Starting and
Ending Points
These experiments use two algorithms to search path from 2a 2b
the three different starting points to the different target point,
of which the results are recorded in Fig.6.

3a 3b
Fig.6 Two algorithms path planning
1a 1b

TABLE IV. TWO ALGORITHMS SEARCH PERFORMANCE

Algorithm Experiment Start point End point Path length Search time Inflection point
A-star 1a （1，1） （15，16） 29 3.554s 5
Improve A-star 1b （1，1） （15，16） 29 0.558s 1
A-star 2a （10，10） （1，20） 19 3.130s 3
Improve A-star 2b （10，10） （1，20） 19 0.507s 1
A-star 3a （20，1） （10，20） 29 3.199s 4
Improve A-star 3b （20，1） （10，20） 29 0.517s 1
It can be seen from Fig.6 that in the three sets of algorithm can find the optimal path each time, and the search
experiments, both the A-star algorithm and the improved A- speed of the improved A-star algorithm is more than 5 times
star algorithm can find the shortest path from the starting point than that of the A-star algorithm. Based on the above
to the target point, but the path of the A-star algorithm experiments, it can be concluded that the improved A-star
searches for has more inflection points. From the path search algorithm not only can search for the optimal path under any
time of the two algorithms in TABLE Ⅳ, under the same search condition, but also searches path much faster than the
conditions, the search speed of the improved A-star algorithm A-star algorithm.
is 5 times faster than that of the A-star algorithm.
IV. CONCLUSION
C. Analysis of Experimental Results To improve A-star algorithm, this paper adds the angle
From the blank gird map search of A, both algorithms can evaluation cost function which reduces the number of
plan the shortest path, but the path that the improved A-star inflection points in the search path and using the jump point
algorithm finds has only one inflection point, which is the search strategy, preprocesses the open list of the A-star
optimal path and the improved A-star algorithm search path 5 algorithm to reduce the number of node searches, thereby
times faster than the A-star algorithm. From the results of the speeding up the search and achieving the goal of quickly
four search results in the experiment a) for the same starting searching for the optimal path. It can be seen from the
and ending position, the improved A-star algorithm can stably experimental results that in any search environment, the
search for the optimal path, while the A-star algorithm can improved A-star algorithm not only has a faster path search
search for the shortest path with the different path trajectory speed than the A-star algorithm, but also can find the optimal
which is not the optimal path. At the same time, the average path. Thus the improved A-star algorithm can find the optimal
search speed of the improved A-star algorithm under the same path more quickly.
condition is more than 5 times than that of the A-star
algorithm. And according to the data of the path planning for REFERENCES
the three groups of different starting points in Experiment b), [1] H.Y Wang, Mobile Robot Industry: Hundreds of Flowers and Rapid
it can be seen that the improved A-star algorithm can search Development ,Logistics Technology and Application,vol.24,pp.66-69,
2019.
for the optimal path from the starting point to the target point
under any conditions, while it is not sure if the A-star

1537

Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 14,2023 at 12:02:46 UTC from IEEE Xplore. Restrictions apply.
[2] J.D. Zhang， Y.J. Feng， F.F. Shi ， et al. Vehicle routing in urban [8] A. Kaplan, N. Kingry , P. Uhing , et al. Time-optimal path planning with
areas based on the oil consumption weight dijkstra algorithm .IET power schedules for a solar-powered ground robot. IEEE Transactions
Intelligent Transport Systems，vol.10,pp.495-502，2016. on Automation Science and Engineering, vol.14, pp. 1235-1244, 2017.
[3] Q.F. Lu ,K. Hancock ,C*: A Bivariate Best First Search to Process [9] S. Wei, L. Wang, B.R. Wang, H.J. Ren, Y.S. Yang, X.L. Liu, Y.C. Ding,
Category Sequence Traversal Queries in a Transportation Network, Improvement of A-star Algorithm and Its Application in AGV Path
IEEE Access，pp.127-136,2010. Planning, Automation Instrumentation,vol.38,pp.51-54, 2017.
[4] B.X. Yu, X.M. Chu, C.G. Liu, H. Zhang, Q.Z. Mao, Using the improved [10] S.H. Zheng, Research on Visual Navigation AGV Positioning and Path
A-star algorithm for unmanned channel survey ship path planning. Planning Technology ,South China University of Technology, 2016.
Journal of Wuhan University (Information Science Edition), v0l.1,2019. [11] D. Harabor, A. Grastien,"The JPS pathfinding system", SOCS, 2012.
[5] A. S, Optimal and efficient path planning for partially-known [12] D. Harabor, A. Botea, P. Kilby,Path symmetries in uniform-cost grid
environments， Robotics and Automation, 1994. Proceedings., 1994 maps, Proceedings of Symposium on Abstraction Reformulation and
IEEE International Conference on, pp.3310-3317,1994. Approximation (SARA), 2011.
[6] J. Lee, B.Y. Kang, D.W. Kim, Fast genetic algorithm for robot path
planning, Electronics Letters, vol.49,pp. 1449-1451,2013.
[7] M.R. Jabbarpour , H. Zarrabi, J.J. Jung, et al. A green ant-based method
for path planning of unmanned ground vehicles. IEEE Access, pp. 1820-
1832,2017.

1538

Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 14,2023 at 12:02:46 UTC from IEEE Xplore. Restrictions apply.