This survey provides an overview of popular pathfinding algorithms and techniques based on graph generation problems. We focus on recent developments and improvements in existing techniques and examine their impact on robotics and the video games industry. We have categorized pathfinding algorithms based on a 2D/3D environment search. The aim of this paper is to provide researchers with a thorough background on the progress made in the last 10 years in this field, summarize the principal techniques, and describe their results. We also give our expectations for future trends in this field and discuss the possibility of using pathfinding techniques in more extensive areas.
Pathfinding is a fundamental component of many important applications in the fields of GPS [
The graph generation problem for “terrain topology” is considered a foundation of robotics and video games applications. In this problem, the pathfinding navigation is conducted in different continuous environments, such as known 2D/3D environments and unknown 2D environments. Several different techniques have been proposed to represent the navigation environment for the graphs of these three scenarios. Each of the representative environment graphs in this paper refers to one of two techniques, skeletonization or cell decomposition [
Skeletonization techniques extract a skeleton from the continuous environment. This skeleton captures the salient topology of the traversable space by defining a graph
There are a number of hypotheses about the properties of the terrain maps produced by skeletonization and cell decomposition as follows.
Scaling the terrain maps will create unnecessary space and variation in the original map.
There is a significant difference between the properties of grid terrain maps (regular and irregular) depending on whether they are for games or robotics.
The design of artificial terrain maps for robotics testing must consider the properties of actual terrain maps.
The second step in the pathfinding process is the search algorithm itself. Here, the problem is to return the optimal path to users in an efficient manner. Games and robotics developers use a variety of techniques.
This paper looks at pathfinding research in the field of video games, with an emphasis on future trends and developments. We summarize recent progress in pathfinding and identify current challenges. Our intention is to inspire researchers and developers and to provide an outlook on both the robotics and video games industries. The relationship between different types of terrain topologies is illustrated in Figure
From left to right: (a) terrain topologies; (b) hierarchical techniques.
A grid is composed of vertices or points that are connected by edges to represent a graph [
Regular grids are one of the most wellknown graph types and are widely used by computer games developers and roboticists. There are a number of video game developers who have worked in this area, producing games such as Dawn of War 1 and 2, Civilization V, and Company of Heroes; roboticists have employed regular grids in the mars rovers spirit and opportunity [
(a) Square grid with three obstacles, (b) hexagonal grid with three obstacles, and (c) triangulation grid with three obstacles and (d) without obstacles.
Square grids are one of the most popular grid graphs in computer games and robotics, and numerous algorithms have been proposed to solve the pathfinding problem for this type of grid (see Figure
Uras et al. [
Harabor and Grastien [
Bnaya et al. [
In terms of changing graphs in the presence of “dynamic obstacles,” Anderson [
Sharon et al. [
Björnsson et al. [
Othman et al. [
Triangular grids (Figure
Nagy [
Unlike the grid graphs discussed above, the cubic grid (Figure
Kiliç and Yalcin [
Recently, Nash and Koenig [
Irregular grids are used in many different applications and fields. In this survey, we highlight all of the wellknown techniques that have been proposed to represent terrain topology.
The visibility graph (Figure
(a) Visibility graph (brown represents obstacles), (b) mesh navigation with triangular grid (blue regions represent obstacles), (c) 3D G2CBS path smoothing: three consecutive points in space from a unique 2D plane (left); planar G2CBS path smoothing is applied to these consecutive points after mapping 3D waypoints to 2D (right) [
To solve the singleagent problem, NaderanTahan and ManzuriShalmani [
For the multiagent problem, Šišlák et al. [
As illustrated in Figure
Šišlák et al. [
In multiagent pathfinding, Kapadia et al. [
Harabor and Botea [
Unlike Kallmann [
Waypoints (Figure
Ferguson and Stentz [
Yang and Sukkarieh [
Lucas et al. [
Cui and Shi [
One disadvantage of techniques that use regular and irregular grids is that they require considerable memory space. Hierarchical techniques mitigate this memory space problem by allowing the continuous environment to be discretized. By applying a finer granularity in certain regions, a more accurate representation can be produced, especially near obstacles. A coarser granularity is applied in regions where detail is not necessary, such as wide, open spaces.
Coulter [
Kamphuis et al. [
Rohrmuller et al. [
Finkel and Bentley [
Reineking et al. [
Hirt et al. [
Naveed et al. [
The singleagent pathfinding search has been studied intensively over different topologies. The main concern for researchers is to provide an optimal path with respect to computation time and memory overhead. Moreover, the best representation of a plane surface is given by a regular grid, specifically a square grid (octile), and uneven land is best represented by an irregular triangular mesh. Drawing artificial maps “created algorithmically” for applications such as video games or robot navigation have different properties than maps created by designers.
Unlike the singleagent case, multiagent pathfinding can take a decoupled or coupled approach. In the decoupled approach, paths are planned for each agent separately. Such algorithms are very fast, but their optimality, and even completeness, is generally not guaranteed. In the coupled approach, the problem becomes a singleagent search that is solved from a higherdimension perspective. One of biggest problems is avoiding collisions between agents during their movement. In this case, we can use the Iterative Taxing Framework (ITF) [
The shortcomings of the above algorithms are that they spend too much time scanning the grid maps and suffer from high memory overheads. Many researchers do not consider the shape of the agents as they navigate from start point to destination. This is especially true when encountering the corners of obstacles and, in the multiagent case, for avoiding other agents.
Navigation through different types of environment can affect the pathfinding results. Most previous work is based on a static and realtime environment, and relatively few works have considered the dynamic case. Singleagent pathfinding generally employs a static environment, because many fields have pathfinding applications for such scenarios. Multiagent cases have been used with realtime and dynamic environments in a few studies. Realtime pathfinding allows the search agent to perform actions while the search is being conducted. This means that partial solutions can be returned, enabling the agent to follow or otherwise incorporate actions into the final solution. The authors are particularly interested in the domain of realtime strategy games, which require path planning for many agents in a common space under the constraints of limited resources and highquality results. A summary of the most widely used techniques for both single and multiagent problems is given in Table
Recently reported pathfinding algorithms used in robotics and video games.
Topologies  Environment “system”  Pathfinding addresses  Exemplification  Memory complexity  Time complexity  Cost metric  Pathfinding technique  Reference 

Undirected uniformcost 
Static  Singleagent  Game development  —  A^{*} + ALTBestp 
ALTBestp, Manhattan, and ALT heuristics  A^{*} and IDA^{*} algorithms  [ 


Undirected uniformcost 
Static  Singleagent  Game development  — 

Manhattan  Improved A^{*} algorithm  [ 


Undirected uniformcost 
Static  Singleagent  Game development  No memory overhead  JPS algorithm 
Manhattan  A^{*}, HPA^{*}, and JPS algorithms  [ 


Undirected uniformcost 
Static  Singleagent  Game development 

—  —  IEA^{*} and IDA^{*} algorithms  [ 


Undirected uniformcost 
Static  Singleagent  Game development  —  SUB algorithm 
—  SUB, BlockA^{*}, CPDfull, CPDmbm, JPSoffline, JPSonline, PDH, PPQ, and Tree  [ 


Undirected uniformcost 
Realtime  Multiagent  Game development  —  —  —  ITF, EITA, and MCITA schemes  [ 


Undirected uniformcost 
Realtime  Multiagent  Game development  —  PRS algorithm 
Manhattan and Euclidean  A^{*}, FS, PBS, and PRS algorithms  [ 


Hexagonal grid  Realtime  Multiagent  Robotics systems  —  —  Euclidean  Augmented A^{*} and Accelerated A^{*} algorithms  [ 


Hexagonal grid  Realtime  Singleagent  Robotics development  —  —  —  D^{*} algorithm  [ 


Triangular grid  Realtime  Multiagent  Robotics and games development  —  —  —  AD^{*} algorithm  [ 


Cubic grid  Static  Singleagent  Robotics and games development  —  —  —  Theta^{*}, Lazy Theta^{*}, and A^{*}  [ 


Mesh navigation  Realtime and dynamic  Multiagent  Robotics and games development  45 KB of memory per agent  —  —  Framework  [ 


Visible graph  Dynamic  Multiagent  Robotics development  AA^{*} less than A^{*}  —  Euclidean  AA^{*} algorithm  [ 


Waypoint  Realtime  Multiagent  Robotics development  —  —  —  Ant colony optimization algorithm  [ 
In this review article, we have summarized recent progress in the field of pathfinding. Basic classes and techniques, which are now in use for pathfinding, have also been discussed in detail. From the literature cited here, it is clear that remarkable efforts had been made to determine accurate realtime paths with minimal or no disturbance to the sample of concern. It is obvious that the basic shortestpath principles in robotics and video games are mature theories, and in coming years we can expect major changes in navigation. Researchers from all over the world are working to improve pathplanning algorithms. One domain that has not been investigated here is augmented reality. This field presents researchers with a range of opportunities and issues, and we believe that the nextgeneration video games industry will be based on the interactive opportunities offered by augmented reality.
The authors declare that there is no conflict of interests regarding the publication of this paper.