A review of robot path planning algorithms

Duan Jingjing*, Liu Jinbo, Wu Yahui, Wang Xiaonan, Cao Liangqiang, Yan Yan, Wang Junjie,
Gong Naifa
Radar Nco School of Air Force Early Warning Academy, Wuhan, Hubei 430019, China
*[email protected]

ABSTRACT

Path planning is an significant research aspect of multi-robot collaboration, which refers to the planning of a suitable
path from the starting point to the target point of multiple robots under the constraint conditions. Path planning
algorithms are divided into classical algorithms and heuristic algorithms, and the principles, advantages and
disadvantages, research progress and so on of each algorithm are introduced. Finally, the existing research is summarized
and prospected, which provides a reference for future research on robot path planning.
Keywords: path planning, Dijkstra algorithm, ant colony algorithm, genetic algorithm

1. INTRODUCTION
With the continuous progress and development of the world economy, more and more countries have realized the
importance of manufacturing in the process of national economic development. Since Germany proposed Industry 4.0,
countries around the world have caught up, committed to improving the competitiveness of their own manufacturing
products, and seize the opportunity in a new round of industrial revolution. This is inseparable from the development of
industrial automation to improve efficiency, optimize resources, and reduce costs. As a big country in the East, China has
made rapid progress in industrial construction since the beginning of reform and opening up, and various industries have
been developed by leaps and bounds. However, in the process of industrialization, it has only roughly caught up with
other developed countries. If you want to overtake a corner and achieve self-breakthrough in a new round of industrial
revolution, you must seize this important opportunity and strive to develop industrial automation and industrial
intelligence. Among them, the research and application of autonomous mobile robot is a very important application.
Autonomous mobile robot has the functions of autonomous perception, decision control, path planning, active
cooperative obstacle avoidance and so on. Among them, the use of multi-robot cooperation, parallel to complete the
complex, multi-step, collaborative path planning problem, which is a hot issue in the current research.
In recent years, China's research in this respect has accumulated, and it has been applied in the market. For example,
many kinds of sorting robots used in logistics warehouses, container handling robots in port terminals, and so on. These
applications greatly save labor costs and improve logistics efficiency. Therefore, at this stage, the research on mobile
robots in the market is also more hot, so as to serve the real hot issues and serve the enterprise and the market. Robot
path planning is also one of the most basic and fundamental problems, which has been discussed and studied by many
experts and scholars.
Path planning is to select an optimal path from the initial position to the target location according to the set preferences
under the specified constraints. Specifically, path planning refers to selecting an optimal or sub-optimal continuous path
from the initial position to the target location according to the objective function, so that the robot can avoid all obstacles
autonomously in the process of movement. Its essence is to obtain the optimal solution or feasible solution under several
constraints. Therefore, the path planning problem mainly achieves three goals: first, planning the feasible path from the
starting point to the target point; Second, there is no collision with any object during the whole path process, and the safe
and smooth driving to the target point; Third, on the basis of feasible and collision-free path, the path is optimized to
meet the corresponding objective function requirements.
According to whether obstacles and node positions change in the working environment, path planning models can be
divided into two categories. Global path planning [1] refers to the planning of the optimal path of the robot from the
starting point to the target point. The whole process only considers the influence of static obstacles and ignores the
dynamic changes. In this case, the global information of the robot's working environment, the position or shape of the
obstacle is fixed, and the position information of the starting point and target node is known and unchanged. This is the

Ninth International Conference on Electromechanical Control Technology and Transportation (ICECTT 2024),
edited by Jinsong Wu, Azanizawati Ma'aram, Proc. of SPIE Vol. 13251, 1325141
© 2024 SPIE · 0277-786X · doi: 10.1117/12.3039469

Proc. of SPIE Vol. 13251 1325141-1

most basic model in the process of robot path planning. Local path planning [2] refers to the robot's real-time collection of
information changes in the environment and timely adjustment of its own driving path to avoid collision during the
process of driving according to the globally planned path. In this case, the position of obstacles is constantly changing,
and the robot needs to detect changes in the surrounding environment through sensors and other means to determine its
own position and the direction of the target node. At this time, the influence of the robot's speed, acceleration and
direction on path planning needs to be considered. In essence, there is little difference between global path planning and
local path planning, both of which bypass obstacles to find the advantageous path to the target point. Many algorithms
that are suitable for global path planning have been improved and can be used for local path planning. The algorithms
can be divided into classical algorithms and heuristic algorithms.

2. CLASSICAL ALGORITHM
The classical algorithm is based on traditional operations research, and the optimal solution of the objective function is
* [4]
obtained step by step by mathematical method. The classical algorithms mainly include Dijkstra[3], A , artificial
potential field method [5] and so on.
2.1 Dijkstra
Dijkstra algorithm belongs to the classical algorithm for solving the path problem. Based on the greedy algorithm, the
algorithm expands from the initial node as the center of the circle, and searches the minimum value of all feasible paths
successively to obtain the shortest path to the destination.
Dijkstra's algorithm is generally used to solve the shortest circuit problem between specified two points, and is currently
considered as the optimal algorithm to solve the shortest circuit problem of the network without negative weight. The
algorithm selects the shortest path node by calculating the path cost of the nodes around the starting node from the
starting node, and then expands outward continuously to determine the shortest path. The basic principle is that if path
v1 , v2 ,..., vn 1 , vn  is the shortest path from v1 to vn , then path v1 , v2 ,..., vn 1 must also be the shortest path from
v1 to vn 1 . The following example is used to calculate the shortest path from v1 to v6 to illustrate the calculation
process of Dijkstra, as shown in Figure 1. During the calculation, mark T  vi  is used as the feasible distance from the
initial point to the point vi , and mark P  vi  is used as the shortest distance from the initial point to the point vi . The
specific steps are as follows:

Step1: Let v1 be the starting point, it can reach node v2 and node v3 , T  v2   4 , T  v3   6 ,Compare the two sizes
T  v2   T  v3  . So make P  v2   4 and record the path v1 , v2  .

Step2: Consider starting from node v2 , which can reach node v4 and node v5 , then

T  v4   P  v2  l24   9
（1）

T  v5   P  v2  l25   8
（2）

T  v3   T  v5   T  v4  P  v3   6 v1 , v3  .
Compare all T , , so let , record path
Step3: Consider starting from node v3 , which can reach node v4 and node v5 , then

T  v4   min T  v4  , P (v3 )  l34   min T  v4  , 6  4  9

（3）

T  v5   min T  v5  , P  v3   l35   min T  v5  , 6  7  8

（4）

Proc. of SPIE Vol. 13251 1325141-2

 T  v5   T  v4  P  v5   8 v , v , v 
Compare all T , , so let , record path 1 2 5 .
Step4: Consider starting from node v5 and reaching the end node v6 , then

P  v6   P  v5   l56  8  5  13
（5）

T  v4   T  v6  v , v , v 
Compare all T , , record path 1 2 4 .
Step5: Consider starting from node v4 and reaching the end node v6 , then

T  v6   min T  v6  , P (v4 )  l46   min T  v6  ,9  7  13

（6）

let P  v6   13 , record path v1 , v2 , v5 , v6 

As can be seen from the above steps, the shortest path from the initial node v1 to the target node v6 is v1 , v2 , v5 , v6  .

v2 5 v4
4 4 7
v1 5 v6
4
6 5
v3 7 v5
Figure 1 Dijkstra.
Dijkstra is an algorithm based on greedy strategy. The algorithm is simple and clear, and it can get the optimal solution.
However, it can not be applied to the network with negative weight, and the complexity is high, and the storage space is
relatively large. However, when the search scope increases, the time required to find the global optimal solution
increases significantly, which is not suitable for real-time search. In view of this shortcoming, based on the traditional
Dijkstra algorithm, the shortest path is searched outward from the initial point and the target point at the same time until
the search results overlap in [6]. This method can obtain the optimal or sub-optimal solution within a limited time. In [7],
the precursor node was introduced when Dijkstra algorithm was used, which not only realized the shortest path planning
of multiple UAVs, but also prevented track conflicts among multiple UAVs. In [8], Dijkstra algorithm was first used for
global path planning, and then combined with PID algorithm for local path planning, which effectively reduced
navigation time and improved the success rate of navigation.
*
2.2 A
A* is essentially the same as Dijkstra, but it adds the guidance of heuristic information, so it gives priority to the path
closest to the target point, shrinks the search area and improves the algorithm speed.

A* set the open table and the close table. The search process starts from the initial node and adds its adjacent nodes to
the open table. By estimating the path cost of adjacent nodes f  n  , the smallest node of f  n  is selected to be
updated as the current node and added to the close table. The process is repeated until the target node is reached, thus
determining the optimal path. The formula for calculating f  n  is as follows.

f n  g n  h n

（7）

Proc. of SPIE Vol. 13251 1325141-3

In the above formula, g  n  refers to the cost of moving from the initial node to the current node n, and h  n  refers to
the cost of moving from the current node n to the target node. The sum of the two, f  n  , refers to the total path cost
through the current node n. The path cost is usually calculated using Manhattan distance, Euclidean distance and
diagonal distance.

A* approaches the destination by searching continuously, traverses all feasible nodes, and obtains the optimal path by
*
calculating the cost of comparison paths. Although A can obtain the global optimal solution theoretically, the search
space increases exponentially with the increase of the search scope and the complexity is high.In [9], the author
*
compares the characteristics of Dijkstra algorithm and algorithm A horizontally and deeply, and the results show that
Dijkstra algorithm can find the shortest distance and obtain the optimal solution, but it is very inefficient for large-scale
problems. The algorithm points the search direction to the target point by means of a heuristic function, which can find
the optimal or sub-optimal solution faster and greatly reduce the time complexity. In [10], according to the application of
* *
algorithm A in multi-robot path planning, it is divided into cooperative algorithm A , hierarchical cooperation
* *
algorithm A and time-window hierarchical cooperation algorithm A .In [11], aiming at the disadvantage that
*
algorithm A increases with the size of the search space, which will lead to serious bottlenecks, A hierarchical path
*
lookup method A is proposed. This method decouples the problem based on grid maps and solves it in two parts: local
cluster and global planning, reducing the complexity of path planning.In [12], it is proposed to add virtual subtarget
*
method to help escape the local minimum trap, and solve the path planning problem of mobile robots by combining A
*
algorithm. In [13], algorithm A is integrated with JPS and security weight square matrix to improve the security and
* *
real-time performance of traditional algorithm A . Both Dijkstra and A conduct search by greedy algorithm, and do
not consider the pros and cons of nodes in the path finding process, which is prone to dimensional disaster, which makes
*
Dijkstra and A limited to solving small and medium-sized problems.
2.3 Artificial potential field method
Artificial potential field method introduces the concept of generalized potential field in physics, and artificially
constructs abstract potential field in vehicle operating environment. Specifically applied to the path planning process,
obstacles generate repulsive force field on the vehicle, and target points generate gravitational field on the vehicle,
making the vehicle avoid obstacles under the action of the combined force and reach the target point. The design idea of
the algorithm is simple and easy to understand, in line with thinking logic, and the calculation amount is smaller than
other algorithms. In addition, the adaptability to complex environments is also good, and can adapt to adjust individual
behavior according to environmental changes. However, the resultant potential field force sometimes has local minimum
value, which leads the algorithm to fall into local optimal. In [14], gradient descent method is used in the artificial
potential field to carry out online path planning, and repulsive force is added to the position during operation to reduce
falling into the local minimum. In [15], a two-stage chaotic optimization method was adopted to obtain the motion state
of the robot through chaotic search, and then the artificial potential field method was improved, effectively overcoming
the disadvantage of unreachable targets. In [16], the influence range of the repulsive force is changed by adding the
influence factor of relative distance, so as to improve the global search capability of the artificial potential field method.

3. HEURISTIC ALGORITHM
The classical algorithm aims to obtain the optimal solution of the problem, but it usually takes a lot of computing time
and space to get the result. In contrast, heuristic algorithm can get the approximate optimal solution or acceptable
satisfactory solution of combinatorial optimization problem efficiently and quickly based on intuitive or empirical
methods at the cost of reducing the accuracy of algorithm. At present, heuristic algorithms are usually derived from
bionics, including ant colony algorithm, genetic algorithm and so on.

Proc. of SPIE Vol. 13251 1325141-4

3.1 Ant colony algorithm
Ant colony algorithm is an intelligent method to solve path planning by simulating the cooperative process of ant
population to find food in nature. It works is to simulate the process of ant groups in nature to find a optimal path from
the ant nest to the food. During the foraging process, ants release a chemical, pheromone, in the path that acts as a
communication mechanism between ant groups. In the same period of time, the ants on the shorter path will return more
times and accumulate more pheromones, which leads the subsequent ants to choose the better path. The group behavior
has a positive feedback effect, and through the feedback mechanism, the group is constantly guided to adjust the path, so
that the solution of the problem develops towards the global optimal direction, and finally the global optimal solution is
obtained.
In practical application, the initial feasible solution of the problem is usually transformed into the key node of the ant
path, and the feasible solution space of the problem to be optimized constitutes all the walking paths of the whole ant
population. In contrast, each ant's path represents a feasible solution. The whole algorithm optimization process is the
result of the greedy strategy and the pheromone strategy, which should consider not only the node with the shortest
distance but also the pheromone concentration. Specifically, each individual ant can change the surrounding environment
by releasing pheromones, and adapt to the environment by sensing changes in surrounding pheromones. Through
pheromones, individuals communicate indirectly, making the search process converge continuously and approach the
optimal solution. Finally, the ant path is transformed into a feasible solution to the problem. The algorithm transforms the
solution space into the key node on the ant path, which is not restricted by the function constraints. The search space
starts with many solutions, that is, a set of feasible solutions, rather than a single solution, and has the capability of
distributed parallel computing, which enables the algorithm to search for comprehensive solutions at multiple points and
effectively reduces the possibility of falling into local optimal solutions. Ant colony algorithm is very good at solving
problems that are difficult to be solved by traditional search algorithms, such as complex nonlinear optimization
problems.
However, the conventional ant colony algorithm [17] is prone to fall into local optimality, prematurity and slow
convergence in the search process. In [18], the Travelling salesman problem (TSP) is mainly studied, and the principles
and methods of ant colony algorithm are comprehensively analyzed. Its effective application to NP-hard problems such
as scheduling problem, assignment problem and set covering problem is also introduced. In [19], a max-min ant system
is proposed, which uses pheromone locus restriction to direct search to high-quality results while preventing premature
convergence. In the aspect of multi-robot path planning, many scholars have also done a lot of research. In [20], adaptive
expectation function is introduced into the transition probability of ant colony algorithm to avoid path devolution by
increasing the gap between the probability of neighboring nodes being selected. The main parameters of the ant colony
algorithm, such as ant population number, heuristic factor, pheromone factor and pheromone volatilization coefficient,
were simulated and analyzed, and the influence of these parameters on the performance of the algorithm and the result
and efficiency of path planning was expounded. An adaptive pheromone update strategy was developed, so that the ant
colony algorithm still had a strong search ability in the later iteration.
3.2 Genetic algorithm
Genetic algorithm is a bionic algorithm that simulates the genetic process of organisms and obtains the dominant
population through multi-generation heredity. It has fast random search ability and strong robustness. The algorithm first
generates the initial population randomly, that is, transforms the feasible solution of the problem into the genes in the
individual chromosomes of the population by coding. Then, the objective function is transformed into the corresponding
individual fitness. Then, through chromosome selection, crossover and variation, fitness function is used to judge the
advantages and disadvantages of new individuals, and excellent offspring are retained and inferior offspring are replaced
to produce new populations. The above operations are repeated until the target number of iterations is reached or the
objective function converges. Genetic algorithm is good at dealing with complex multidimensional problems, can search
the neighborhood quickly, has good algorithm convergence, and is relatively easy to get the global optimal solution.
Secondly, the algorithm is robust and very suitable for parallel processing, which enables multiple ants to search
solutions simultaneously. Moreover, the genetic algorithm is extensible. According to the defects of its own algorithm,
the genetic algorithm is combined with the complementary algorithm to obtain a compound algorithm with better
performance. In addition, the algorithm can also add prior knowledge to improve the search efficiency.
The genetic algorithm also has some disadvantages, such as local optimization when individuals converge too early.
Often get inaccurate planning results; The process of parameter fine-tuning is complicated. In [21], the lack of heuristic

Proc. of SPIE Vol. 13251 1325141-5

rules in the mutation operator of the traditional genetic algorithm is improved to promote the genetic algorithm to quickly
find the advantage path in a dynamic environment. In [22], insertion operators and deletion operators are added in the
genetic process, so that the robot can obtain an optimized barrier-free path.

4. SUMMARY AND PROSPECT

This paper summarizes common robot path planning algorithms, which can be divided into two categories according to
their application scenarios, global path planning algorithms and local path planning algorithms. No matter what kind of
application scenario or situation, the application of the algorithm is connected, integrated, and can be mutually universal.
At this stage, the research mainly includes Dijkstra algorithm, algorithm and artificial potential field method in classical
algorithms, and ant colony algorithm and genetic algorithm in intelligent optimization algorithms. These algorithms
represent two kinds of problem solving ideas, which have their own advantages and limitations in the application process,
so they have different application scenarios. Can not only say the shortcomings of a certain algorithm, to look at the
overall level, each algorithm has its own advantages and disadvantages, can not be generalized. Therefore, in view of the
current research, the author puts forward the following prospects:
(1) The development and application of deep learning, such as Q-learning and neural network algorithms, have very
important application significance in improving the visual perception of robots and the accuracy of postural movement
turning, etc., and are more adaptable to complex and changeable environments. These algorithms are also more accurate
and are the frontier academic direction of many researches. In the follow-up research process, researchers should
combine this aspect to conduct multiple exploration and research. The research will be more inclined to the study of this
imitation wheat, on the one hand, it can improve the speed and efficiency of the algorithm, on the other hand, the
algorithm will be more recognized, and the future scalability will be stronger and more extensive.
(2) Because of its own inherent attributes, a single algorithm has some insurmountable shortcomings in the process of
algorithm formation, although the shortcomings of each algorithm are different. According to many existing studies, the
integration of multiple algorithms can make up for each other's shortcomings and increase their strengths, which can
effectively overcome shortcomings and improve the overall search capability, robustness and reliability. However, in the
process of fusion, it may also lead to some problems, such as the complexity of the fusion algorithm will increase greatly,
the real-time responsiveness will become worse, and the difficulty of the subsequent improvement and promotion of the
algorithm will increase. Therefore, in the process of algorithm fusion, we must selectively combine algorithms to reduce
algorithm complexity on the basis of combining the advantages of each algorithm as far as possible.
(3) Robot movement involves specific requirements in many aspects, such as information perception, intelligent decision
making, path planning and motion control. In the process of research, if the path planning is considered separately, the
problem will become simpler and easier to focus on the development of the problem. However, from the perspective of
practical application, it is unrealistic to only consider the path planning research, so it is necessary to be familiar with the
frontier content of other aspects and conduct combined research involving the hardware conditions, input and output of
path planning, so as to provide effective reference for the research.

REFERENCES

[1] Zhang Yalong, Xiao Yinbao. Research and Application of Vehicle Routing Problem Based on Intelligent
Optimization Algorithm [J]. Science and Technology Wind, 2023(36): 10-12.
[2] Ou Jinjun, Zhu Feng. A Collision avoidance planning Method for Multi-Mobile Robots [J]. Robot, 2000(06):
474-481.
[3] Dan D Z, Qing J S. A New Algorithm Based on Dijkstra for Vehicle Path Planning Considering Intersection
Attribute [J]. IEEE ACCESS, 2021, 9 19761-19775.
[4] Shi J ,Su Y, Bu C, et al. A mobile robot path planning algorithm based on improved A [J]. Journal of Physics:
Conference Series, 2020, 1486 (3):
[5] Kovács B, Szayer G, Tajti F, et al. A novel potential field method for path planning of mobile robots by
adapting animal motion attributes. [J]. Robotics and Autonomous Systems, 2016, 82 24-34.
[6] Noto M, Sato H. A method for the shortest path search by extended Dijkstra algorithm[C]//IEEE International
Conference on Systems.IEEE, 2000, pp. 2316-2320 vol.3.

Proc. of SPIE Vol. 13251 1325141-6

[7] Huang Yihu, Yu Yanan. Research on anti-conflict shortest path Planning based on improved Dijkstra Algorithm
[J]. Computer and Modernization, 2022(08): 20-24.
[8] Song Jiangyi, Li Dan, Chen Wenbo. Local path planning of indoor mobile Robot based on Dijkstra and PID [J].
Journal of Anhui University of Technology (Natural Science Edition), 2023,01(40): 59-64.
[9] Soltani R A, Tawfik H ,Goulermas Y J , et al. Path planning in construction sites: performance evaluation of
the Dijkstra, A*, and GA search algorithms. [J]. Advanced Engineering Informatics, 2002, 16 (4): 291-303.
[10] SILVER D. Cooperative Pathfinding[C]//. Proceedings of the First Artificial Intelligence and Interactive Digital
Entertainment Conference, June 1-5, 2005, Marina del Rey, California, USA, 2005.
[11] Botea A, Martin Müller, Schaeffer J .Near optimal hierarchical path-finding (HPA*)[J].journal of game
development, 2004,1.
[12] Wang Hongbin, Hao Ce, Zhang Ping, et al. Path Planning of Mobile Robot Based on A~* Algorithm and
Artificial Potential Field Method [J]. China Mechanical Engineering, 2019, 30 (20): 2489-2496.
[13] Zhi Chenbo, Zhang Aijun, Du Xinyang, et al. Research on global path Planning of Mobile Robot with improved
A* Algorithm [J]. Computer Simulation, 2023,40(2): 486-491.
[14] Bounini F , Gingras D , Pollart H ,et al.Modified Artificial Potential Field Method for Online Path Planning
Applications[C]//Intelligent Vehicles Symposium.IEEE, 2017.
[15] Chen W ,Wu X ,Lu Y . An Improved Path Planning Method Based on Artificial Potential Field for a Mobile
Robot [J]. Cybernetics and Information Technologies, 2015, 15 (2): 181-191.
[16] Liang Bin, Yin Yifei, Xu Ming, et al. Obstacle Avoidance Algorithm for AGV Intelligent Vehicle Based on
Improved Artificial Potential Field Method [C]//. The 17th National Conference on Signal and Intelligent
Information Processing and Application, Chongqing, China, 2023.
[17] Liu Chang 'an, YAN Xiaohu, LIU Chunyang, et al. Dynamic Path planning Method of Mobile Robot based on
improved ant Colony Algorithm [J]. Acta Electronica Sinica, 2011, 39 (05): 1220-1224.
[18] Dorigo M ,T. Stützle. The ant colony optimization metaheuristic: Algorithms[J]. new ideas in optimization,
2003.6(16-20):187-191.
[19] Thomas Stützle, Hoos H H. MAX-MIN Ant system[J]. Future Generation Computer Systems, 2000,16(8): 889-
914.
[20] Du Lei, Shi Jianfang. Research on path planning of mobile robot based on improved ant colony algorithm [J].
Journal of Taiyuan University of Technology, 2019,50(4): 523-528. (in Chinese)
[21] Liu Erhui. Path Planning of automatic guided Vehicle Based on Improved Genetic Algorithm and its
Implementation Platform [J]. Computer Integrated Manufacturing Systems, 2017, 23 (03): 465-472.
[22] Pei Yijian, Yang Liangliang, Yang Chaojie. Path planning of Mobile Robot based on a hybrid Genetic
Algorithm [J]. Modern Electronics Technique, 2019, 42 (02): 183-186.

Proc. of SPIE Vol. 13251 1325141-7