Semi-Autonomous Navigation for Virtual Tactical Simulations in the

Military Domain

Juliana R. Brondani

, Luis A. L. Silva

, Mateus B. Rutzig

, Cesar T. Pozzer

, Raul C. Nunes

João B. Martins

and Edison P. de Freitas

Graduate Program in Electrical Engineering, Federal University of Rio Grande do Sul, Porto Alegre, RS, Brazil

Graduate Program in Computer Science, Federal University of Santa Maria, Santa Maria, RS, Brazil

Keywords: Semi-autonomous Navigation, Tactical Simulation.

Abstract: Integrated constructive and virtual simulations are becoming popular for tactical training in the military do-

main. An important aspect concerning the integration of these simulation models in the construction of virtual

tactical simulations is the modelling and implementation of different kinds of semi-autonomous agents. A

fundamental feature of these agents is the capability of intelligently and realistically modelling task-oriented

navigation activities in large virtual terrain simulation environments, while following underlying military

doctrine and tactics. This paper reviews important navigation issues that emerge in such simulation systems

and prominent Artificial Intelligence (AI) techniques that have been explored to solve them. From this anal-

ysis, a hybrid, semi-autonomous navigation framework is proposed aiming to fulfil the needs of virtual tactical

training simulations, more specifically, in the military domain. As implemented in a system for the virtual

tactical simulation of artillery battery tasks, the framework shows how to overcome the challenges of imple-

menting realistic global and local navigation behaviours for military units and, at the same time, it shows that

the semi-autonomous behaviours implemented are of primary importance to allow interaction with users for

learning purposes in the simulation exercises.

1 INTRODUCTION

Simulation systems are gaining more popularity for

educational purposes, in which one of the most tradi-

tional and prominent scenarios is the case of military

training (e.g. (Heinze et al., 2002); (Fletcher, 2009)),

although there is an increasing interest in other do-

mains, like industrial (e.g. (Murphy and Perera,

2002)) and medical (e.g. (McGaghie et al., 2010)) to

name a few. These simulation systems present them-

selves as challenging environments for the investiga-

tion and proposition of solutions for semi-autono-

mous navigation problems as they allow agents to

recreate different behaviours realistically.

In military simulations, Computer Generated

Forces (CGFs) or Semi-Autonomous Forces (SAFs)

(Tambe, Johnson et al., 1995) are developed to popu-

late constructive, virtual and blended simulation sys-

tems (Stevens et al., 2015). It is expected that these

forces would act as substitutes for key real-life enti-

ties, for which semi-autonomous navigation behav-

iours stand out when users need to be in the control of

the simulated entities as part of the development of

different military training goals. For this reason, the

primary goal of these agents is to replicate relevant

aspects of human behaviours realistically, while fol-

lowing military doctrine. When dealing with the im-

plementation of semi-autonomous navigation behav-

iours, there are intelligent techniques proposed in the

literature. However, most of these techniques only

handle part of the relevant navigation issues in these

systems, which indicates that these techniques seldom

consider the implementation of complex, layered sim-

ulation scenarios. As discussed in this paper, there are

dynamic navigation and collision detection/avoid-

ance issues to be considered by a combination of

global and local agent navigation behaviours. These

behaviours should be consistent with real-life agent’

actions. Moreover, for any military simulation, agents

also compute navigation actions relying on data struc-

tures that represent relevant characteristics of large

real-life terrain environments.

From this landscape, this paper discusses alterna-

tive navigation techniques ought to be combined in a

Hybrid, Semi-Autonomous Navigation Framework to

Brondani, J., Silva, L., Rutzig, M., Pozzer, C., Nunes, R., Martins, J. and Freitas, E.

Semi-Autonomous Navigation for Virtual Tactical Simulations in the Military Domain.

DOI: 10.5220/0006863104430450

In Proceedings of 8th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH 2018), pages 443-450

ISBN: 978-989-758-323-0

443

support the development of realistic simulations in

virtual tactical simulation systems. As proposed here,

these virtual tactical simulations provide an environ-

ment in which low- or mid-level decision makers can

train their skills in how to better move and employ

their units in a battle scenario to accomplish their mis-

sions. Filling the gap that exists in between construc-

tive and virtual simulations, this type of virtual tacti-

cal simulation is grounded on a realistic scenario that

presents all the necessary elements to exercise the tac-

tical skills of these low- or mid-level commanders.

The paper is organized as follows: First, promi-

nent techniques used to address the agents’ navigation

tasks are analyzed. Second, the virtual tactical simu-

lation is discussed, highlighting its main characteris-

tics and differences from the other types of simula-

tion. Third, the Hybrid Semi-Autonomous Naviga-

tion Framework exploring the studied concept of vir-

tual tactical simulation is presented and analyzed

along with an application example. Then, discussions

are presented highlighting relevant aspects raised dur-

ing the development of the proposed framework, and

finally, the concluding remarks are presented.

2 AGENT’S NAVIGATION

BEHAVIORS IN SIMULATION

SYSTEMS

Navigation issues have been handled by different

techniques proposed in the literature (Botea et al.,

2013) (Kapadia and Badler, 2013) (Algfoor et al.,

2015). This section describes the most used ones in

simulation systems for the military domain, allowing

one to assess whether they cover the semi-autono-

mous navigation requirements of agents involved in

virtual tactical simulation environments.

2.1 A* Algorithms

The A* algorithm (Nilsson, 1998) is an instance of a

deliberative approach to the solution of navigation

problems. Although the A* algorithm is widely ex-

plored in different fields, the computational cost of

executing it may become prohibitive. That is because

the computation cost of the A* sharply increases as

the size of the virtual simulation environment and the

number of agents increase. Some proposals to deal

with known limitations of the A* algorithm are worth

mentioning: i) The Local Repair A* (LRA*) de-

scribes a family of algorithms based on the recalcula-

tion of the remaining of an agent route when a colli-

sion with other simulation objects is imminent (Sil-

ver, 2005). ii) The Cooperative A* (CA*) searches

for a path while considering the routes which are

planned for other agents being executed in the simu-

lation system (Silver, 2005). iii) The D* algorithm

(Stentz, 1994) is capable of planning paths in un-

known or partially known and changing environ-

ments as it is dynamic in the sense that the cost pa-

rameters used can change during the search for the

solution. This technique also has known optimiza-

tions called D* Lite (Koenig and Likhachev, 2002)

and Anytime D*(Likhachev et al., 2005). However,

these proposals still have computational limitations

when the virtual terrain size increases as this situation

is often related to the increasing of the memory space

requirements and the complexity of the simulation

scenario. In these cases, hierarchical A* techniques

are promising as they speed up the pathfinding pro-

cess by reducing the complexity of the problem sce-

nario as the pathfinding is broken down in a hierar-

chic structure (Cui and Shi, 2011). One of the first

proposals of this hierarchical algorithm (Botea et al.,

2004) proposes the HPA* (Hierarchical Path-Finding

A*), a “hierarchical approach for reducing problem

complexity in path-finding on grid-based maps.” This

technique proposes the creation of abstraction levels

relying on clusters obtained from a regular grid ter-

rain representation. The hierarchical approach miti-

gates the memory space problem and allows faster re-

sults to be computed. However, when dealing with a

virtual terrain with large dimensions, as it is often the

case of a simulation system for representing a real-

life environment, it is not feasible to use a terrain rep-

resentation, which is based on a regular grid. (Dooms,

2013) uses an A* adaptation to navigate upon a quad-

tree representation of the terrain, allowing rapid ad-

aptation to a terrain representation which can be dy-

namic as this solution only requires a limited number

of nodes to be re-evaluated. This algorithm still pre-

sents some drawbacks like the absence of parallelism

during the search and the use of a regular grid in the

representation of the space that is inside of each quad-

tree node.

2.2 Potential Fields

Potential fields and influence maps are strategies

commonly used to treat dynamic obstacles (Silveira

et al., 2010) (Hagelbäck, 2012). They are concepts

originated in the robotics fields, first introduced for

the treatment of real-time obstacle avoidance, which

is a central issue in the development of navigation al-

gorithms for mobile robots (Khatib, 1986). The main

SIMULTECH 2018 - 8th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

444

idea of the potential fields approach is to place attract-

ing or repelling charges at points of interest in a sim-

ulation map. The overall idea is that the agent can cal-

culate the resulting force according to the fields that

are available in the positions around the agent, and

then navigate to the most attracting position in the

near surroundings. The problem of this reactive be-

haviour is that the navigation algorithm may get stuck

at local optima where the highest potential position is

the current position of the agent, but this highest po-

tential position is not the destination position. In

(Hagelbäck, 2012), this problem was handled by as-

signing small repelling fields to the last agent posi-

tions, like a pheromone trail used by ants. This ap-

proach was successfully used in the Open Real Time

Strategy (ORTS) system (Hagelbäck and Johansson,

2008) as the maps representations used present large

open areas. As reported in (Hagelbäck, 2012), in

which maps from the popular RTS game StarCraft

were considered, which are complex and have many

choke points, the solution proposed did not work well

due to a large number of local maxima. To overcome

these problems, a potential field technique combined

to a traditional A* was used, resulting in a hybrid ap-

proach. Although this global-local approach gener-

ates a more realistic behaviour for each agent, it in-

curs in a high computation cost because the agent

keeps analyzing its surroundings and calculating the

potential of each cell in its local map representation.

2.3 Steering Behaviours

(Reynolds, 1999) defined steering behaviour for au-

tonomous agents as the ability to navigate around

their world in a life-like and improvisational manner.

Different steering behaviours like seek, flee and ar-

rive for a simple vehicle model are proposed. Then

these basic behaviours can be combined as to gener-

ate ones that are more complex. The steering behav-

iours are described in terms of the geometric calcula-

tion of a vector representing the desired steering force

(Reynolds, 1999). Due to the representation simplic-

ity of this approach in the solution of navigation prob-

lems, it is a popular framework in the computer game

scenario. Despite this fact, problems like a trajectory

with oscillations or a resultant vector being zero can

appear when the steering approach is used. (Frey,

2015) describes attempts to overcome these steering

difficulties with the introduction of weighting, prior-

itization and awareness concepts. However, these at-

tempts might not solve the cited problems or, if they

do, they increase the complexity of the system. For

this reason, a technique called Context Steering is

proposed. The basic idea is the use of context maps,

where these maps describe interest and danger vector-

based points around an agent. Then, map information

is combined to elect which vector describes the better

final decision according to the current state of the sys-

tem. This resulting approach maintains the benefits of

the steering behaviour since it creates a more “intelli-

gent” behaviour at a local level for each agent.

2.4 Velocity Obstacles

Velocity Obstacle (VO) is most commonly used in

the solution of reactive or local navigation problems.

This technique emerged in the field of robotics (Fio-

rini and Shiller, 1998). VO defines a set of robot ve-

locities (in geometric representation) that would re-

sult in a collision between the robot and a moving ob-

stacle. However, the VO approach presents some

drawbacks. One of them is the high computation cost

to keep updating the velocities of agents during the

execution of the system. Another difficulty is that

agents do not consider that other agents also have de-

cision-making abilities. In (Van den Berg et al.,

2008), this issue is addressed with the development of

the Reciprocal Velocity Obstacle (RVO) approach.

RVO assumes that other agents in the simulations are

also capable of similar collision-avoidance reasoning

while guaranteeing oscillation-free motions. How-

ever, RVO ended up creating another issue called “re-

ciprocal dance” (Feurtey, 2000). Reciprocal dance

occurs when two agents cannot reach an agreement

on which side to navigate around each other. To over-

come this problem, a Hybrid Reciprocal Velocity Ob-

stacle (HRVO) (Van Den Berg et al., 2011) approach

was proposed. HRVO eliminates almost all oscilla-

tions in practice, but it does not guarantee smoothness

in the movement. Therefore, trying to address all the

problems cited before, the Optimal Reciprocal Colli-

sion Avoidance (ORCA) (Snape et al., 2012) can the-

oretically guarantee no oscillations and smooth

movement.

3 VIRTUAL TACTICAL

SIMULATIONS: FILLING

THE GAP BETWEEN

CONSTRUCTIVE AND

VIRTUAL SIMULATION

SYSTEMS

There is a live discussion about military simulation

systems and their applicability. A consensus that

emerges among the practitioners in this area is that

Semi-Autonomous Navigation for Virtual Tactical Simulations in the Military Domain

445

there are clear benefits of using live, virtual and con-

structive simulations for military training. Such ben-

efits are even augmented when these three conceptual

types of simulation are combined in joint LVC simu-

lations (Hodson and Hill, 2014). Despite these con-

ceptual classifications, there are intermediary zones

between constructive and virtual, and between virtual

and live simulations (Meyer et al., 2014). The use of

virtual simulations usually targets the training of tech-

nical and operational skills regarding a given military

equipment. Thus, this specific application of virtual

simulations can be characterized as virtual technical

simulations. The second application of virtual simu-

lations, which is more linked to constructive ones,

does not target the training of how to handle a piece

of equipment, but it is concerned with tactical aspects.

This usage contrasts to constructive simulations,

which are concerned with higher-level (strategic) as-

pects, completely abstracting finer details of a spe-

cific terrain, for instance. Thus, it is possible to define

this second kind of virtual simulations as virtual tac-

tical simulations.

Considering its proximity to the constructive sim-

ulation, the best way to build-up this virtual tactical

simulation concept is to present examples of situa-

tions in which this type of simulation is used. In our

project, the employment of an artillery battery in a

military operation can be taken as a motivational ex-

ample. In this simulation application, the role of a

constructive simulation is to train high-ranked com-

manders in the situations in which it is worth using

the battery. In this type of simulation, higher-ranked

officers (e.g., generals and their high-level staff) se-

lect priority areas and targets in which the battery will

act, for instance. In this constructive simulation, how-

ever, the engagement of desired targets occurs ac-

cording to given rules, and this is enough to provide

automatically generated results to be analyzed by

these high-ranked commanders. Everything in be-

tween from the moment of the decision-making to the

return of the simulation results is entirely abstracted,

i.e., without any interference of the intended users.

Analyzing what happens between these above de-

scribed moments, several tactical decisions about

how to better employ the battery are taken in a real-

life military operation. From the moment in which

commanders of the battery receive the command to

engage a given set of targets, low- or mid-level mili-

tary personnel also need to tactically analyze the ter-

rain, to select where the most suitable routes are as to

safely navigate with vehicles that compose the bat-

tery, among other doctrine-based tasks. It is relevant

to observe that a virtual technical simulation system

is not able to train someone with the necessary skills

to take these tactical decisions. However, where will

the commander of the battery train his/her tactical

skills to best select and execute a set of tactical actions

as mentioned above? It is particularly for this type of

real-life training situation that the virtual simulation

is tailored for.

In virtual tactical simulations, there are different

degrees of user interaction, which go from very high-

level orders (like in the constructive ones) to more de-

tailed ones (which are closer to technical-virtual or

life simulations). Depending on the situation being

simulated, these levels of interaction, which need to

be reflected in the levels of autonomy of the simula-

tion algorithms being used, may also be used alter-

nately. In this context, an important issue for virtual

tactical simulation systems in the field of military

training is that they are used for educational purposes.

In these systems, therefore, simulated agents have the

capacity of receiving both computational and human

inputs, as well as the capacity of reasoning about the

best outcome of these actions according to their cur-

rent situation.

4 A CASE FOR A HYBRID

SEMI-AUTONOMOUS

NAVIGATION FRAMEWORK

FOR VIRTUAL TACTICAL

SIMULATIONS

Navigation algorithms described in the literature are

mostly focused on solving specific navigation prob-

lems. It means that they are rarely combined to handle

realistic simulation scenarios where the navigation is-

sues change dynamically. To address this drawback,

a Hybrid, Semi-Autonomous Navigation Framework

is proposed aiming to fulfil the needs of virtual tacti-

cal training military simulations. The overall idea of

the proposed hybrid navigation framework is to com-

bine global and local navigation approaches to tackle

static and dynamic simulation issues over open fields

and roads of a large virtual terrain considering the

doctrine-oriented navigation demands of an artillery

battery in a tactical battle scenario.

The global navigation approach works over a

symbolic representation of the virtual terrain environ-

ment to search longer paths where the algorithms

used have to consider simulation performance con-

straints. To couple with that, the proposed framework

employs a multi-resolution quadtree data structure

that symbolically represents static obstacles found in

the virtual terrain scenario (Figure 1 (A)). In many

senses, such multi-resolution quadtree supports the

SIMULTECH 2018 - 8th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

446

use of a hierarchical A* navigation algorithm. Thus it

is possible to find paths in large virtual terrains with

lower overhead than traditional A* algorithm.

In Figure 1 (A), a real large terrain (50km²) is di-

vided into a quadtree containing thirteen representa-

tion levels. The proposed quadtree coupled to the hi-

erarchical A* automatically adjusts the depth of the

structure considering non-functional requirements of

the simulation system, such as available computation

capability and memory footprint. For instance, if

there are enough memory and execution time for a

finer search of an agent route in the virtual terrain en-

vironment, the algorithm considers finer terrain reso-

lutions to allow the construction of a more accurate

navigation result. Based on the division shown in Fig-

ure 1 (A), the representation structure can reach a res-

olution around 2 meters in its finest level, allowing

the representation of relevant obstacles such as rivers,

mountains and trees, as these obstacles are identified

in real-life terrain regions of military training (i.e., us-

ing the different map representation of a military

training field). As far as static navigation obstacles

are concerned, the multi-resolution terrain represen-

tations allow the algorithm to quickly detect a path

from one point to another in the virtual terrain if such

path exists. When constructing such quadtree struc-

ture upon the terrain, the navigation algorithm easily

identifies all the navigation-prone neighbouring

nodes of the one that is being analyzed (Figure 1 (B)).

As a result, a path is returned (Figure 1 (C) as required

to simulate a tactical action involving a path-follow-

ing movement of an agent in a simulation exercise –

such kind of path is represented by the line having

circle markers in this Figure 1).

Figure 1: (A) Example of the terrain represented in a quad-

tree structure. (B) Connections between the nodes of the

quadtree. (C) Path defined by the A* upon the quadtree rep-

resentation.

Overtaking dynamic obstacles while the simula-

tions are running is not a straightforward task. Con-

stantly recalculation of paths for every agent may be

prohibitive during real-time simulation executions

even when an optimized virtual structure is available,

such as a quadtree, to also represent the whereabouts

of dynamic obstacles. For this reason, a local naviga-

tion strategy needs to be used in combination with a

global pathfinding algorithm. Unlike the global navi-

gation, this local navigation perspective considers

that agents do not have a global view of the simula-

tion environments since they just maintain a continu-

ous and reactive relationship with their local sur-

roundings. Importantly, such local strategy is respon-

sible for the detection of dynamic obstacles and other

dangers that an agent may face during the virtual tac-

tical simulation exercises. To do so, the proposed

framework explores steering behaviour techniques

that are based on mathematical representations of the

forces in the modelling and implementation of local

navigation behaviours. Among other reasons, these

algorithms allow a quick calculation of the resulting

steering force, which is then used in the agents’ local

navigation actions.

As previously described, besides solving the

global and local navigation, the framework needs to

consider the military tactical doctrine to perform

these agents’ movements, as this requirement is fun-

damental to implement intelligent navigation algo-

rithms in virtual tactical simulation settings realisti-

cally. During the movement of a battery, although

agents adapt according to the current battle and ter-

rain situations, they are implemented as to not deviate

too much from a formation-kind of convoy organiza-

tion detailed by the doctrine (e.g., to maintain a col-

umn formation while moving). Different from many

computer game implementations, the proposed navi-

gation framework considers that the navigation guid-

ance coming from the military doctrine has a higher

priority than the global and local navigation algo-

rithms being executed.

5 THE SOLUTION OF A

PRACTICAL SIMULATION

SITUATION FOR GLOBAL AND

LOCAL SEMI-AUTONOMOUS

NAVIGATION

A practical simulation situation of a concrete military

exercise involving an artillery battery is illustrated in

Figure 2. This situation is representative of others that

occur when virtual tactical simulations are developed

Semi-Autonomous Navigation for Virtual Tactical Simulations in the Military Domain

447

in the implemented simulation system. As described

here, this situation illustrates the use of the global and

local semi-autonomous navigation algorithms imple-

mented in the framework. In doing so, this simulation

case involves a convoy with five vehicles parked at

point A as shown in Figure 2. According to the tacti-

cal goals of an artillery battery mission, the convoy

needs:

• to move from A to B;

• to deploy its artillery units in tactical positions in

B according to the military doctrine.

For the first goal, the hierarchical A* algorithm de-

fines, over the multi-resolution structure of the virtual

terrain representation, the best path from A to B (rep-

resented by dotted line in Figure 2). The best path is

not only the shortest path between these points, but

also the path that can consider key terrain character-

istics and the other forces (i.e. other agents) being

simulated in the current battle situation. It is im-

portant to notice that the defined path which is gener-

ated by the pathfinding algorithm considers move-

ment over different types of scenarios: roads and open

fields. When there are roads in the area (from A to C),

the pathfinding algorithm computes the route over an

undirected graph that contains information about the

roads and their interconnections. However, when an

open field area is considered in the movement (from

C to B), the pathfinding algorithm computes the route

according to the obstacle information present in the

nodes of the aforementioned quadtree structure. In the

end, this quadtree structure represents the information

that is observed by the agents allowing them to move

over the virtual terrain representation.

Figure 2: Example of a convoy execution a navigation task.

During the navigation of the convoy towards B,

the vehicles may face situations that were not ex-

pected when the A* algorithm was executed (i.e.,

when the path was calculated). During the execution

of the simulation exercise, as illustrated in Figure 2,

the first two vehicles of the convoy crossed the bridge

that connects A to B (passing by C). However, the

third vehicle broke the bridge and fell into the river as

a result of such a non-deterministic simulation situa-

tion. When this happens in this simulation system, the

simulation stops since the route from A to B as

planned by the global navigation algorithm no longer

exists. As defined in this virtual tactical simulation

system, this is a relevant military problem which was

selected to be simulated due to its educational value

to the users, among other such problems treated by

the system. In summary, such kinds of problems offer

relevant opportunities for simulation-based training.

Different from fully-autonomous simulation scenar-

ios, which are more common in constructive simula-

tion systems, the users here are invited to decide

which course of action they need to take according to

their knowledge and experience with the underlying

military doctrine and their mission goals, under these

new conditions. One course of action is to allow the

vehicles which already crossed the bridge (vehicles 1

and 2) to continue over the original route to reach the

destination B. When this happens, the hierarchical A*

algorithm is called again as to compute a new route

from D to B for the remaining vehicles (vehicles 4

and 5). This way, a new route from the D to B (repre-

sented by the solid line in Figure 2) is created to allow

the movement of vehicles 4 and 5. It is important to

notice the split of the original convoy and the need for

the user interaction to decide what the simulation sys-

tem should do next, which reveals that a completely

autonomous solution is not suitable in this kind of

simulated situation. As this fragment of simulation

exercise shows, there is a need for a certain degree of

agent semi-autonomy, which is provided by the pro-

posed framework that was implemented in the simu-

lator. Similar to this one, other situations also require

the intervention of users as the tactical training goals

of the simulation system requires. So, this simulation

training scenario requires the implementation of other

degrees of semi-autonomy in the global and local

agent navigation behaviours as this is likely to be the

case of other simulation systems similar to this one.

For the second goal, the convoy is deployed in a

selected region in the virtual terrain scenario accord-

ing to the rules of the tactical military doctrine. In the

example illustrated in Figure 2 (zoomed portion on

the top-left), when vehicles approach their tactical po-

sitions, many conflict points (CP) are found, as it is

expected to be the case of other multi-agent simula-

tion systems. It means that the A* algorithm was used

to define paths for different agents, where these paths

SIMULTECH 2018 - 8th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

448

produced overlapping routes, which may cause colli-

sions among the agents while the simulations are run-

ning. In these dynamic simulation situations, the local

navigation algorithm takes over as implemented in

the simulator, applying steering behaviour forces to

avoid collision between dynamic obstacles. In the im-

plementation of social agent rules to solve such con-

flict situations, such as the rule: stop the movement

and give preference to, for instance, the switch be-

tween the global and local navigation does not need

to have user interaction, thus emphasizing the im-

portance of combining both navigation types in a hy-

brid solution. Even when these local algorithms are in

the control of the agents’ actions, it is relevant to no-

tice that the implemented algorithms also need to con-

sider that the guidelines of a military doctrine have to

be followed; otherwise, the local movement actions

will look like randomly implemented, which is some-

thing that provides a poor realism for the simulations.

While in many kinds of computer games the realism

may not be an issue while dealing with local move-

ment actions, this is a relevant aspect implemented in

our simulation system. As described here, parameters

in the semi-autonomous local algorithms (e.g., veloc-

ity, direction, etc.) are also explored by users under

training to allow them to guide these local actions of

movement as to fulfil defined training goals of the

tactical simulation exercises.

6 DISCUSSION

In the context of military tactical training, it happens

to be impractical to recreate complete scenarios using

real people and equipment due to the cost, amount of

required resources and even the danger in the han-

dling of military equipment without making sure that

users achieved a certain level of training maturity. In

this setting, constructive and virtual simulations have

been used for personnel training by familiarizing the

trainees with tools, vehicles, equipment, military doc-

trines and routines employed in real life. It also means

that there has been an increase of interest in the com-

bination of different types of simulations leading to

the term “blended training” (Stevens et al., 2015). Ex-

amples in the military simulation field that use

blended architectures integrating a constructive sys-

tem with human-in-the-loop are the Royal Australian

Force (RAAF) simulation system (Heinze et al.,

2002) and the semi-automated forces (SAFORS) in

the large-scale SIMNET environment (Tambe et al.,

1995). Different from these approaches, the proposal

presented in this paper is a hybrid semi-autonomous

navigation framework that allows a customized con-

trol of the tasks and agents being simulated. In addi-

tion, in virtual tactical simulations, the use of semi-

autonomous algorithms become relevant as they com-

bine a constructive navigation behavior with the pos-

sibility of human interference, which can provide dif-

ferent benefits for trainees involved in the simulation

exercises.

In the field of simulation, a hybrid solution is also

proposed in (Sahli and Moulin, 2005), describing how

to support wildland fire suppression actions in a vir-

tual environment scenario. As described in their

work, the single-use deliberative techniques (like A*)

is not sufficient to solve real problems, as their appli-

cation problem can present too many constraints.

These reasons also sustain the proposal of hybrid nav-

igation algorithms to navigate in simulation environ-

ments. The use of multiple navigation techniques al-

lows agents to plan a safe and fast route using a global

planner. It also allows such agents to deal with dy-

namic obstacles, other agents, and unforeseen situa-

tions by using local navigation techniques.

7 FINAL REMARKS

The modeling and implementation of realistic naviga-

tion behaviors is a fundamental feature for intelligent

agents in military simulation systems. This paper re-

visits the most prominent solutions in this area, high-

lighting key aspects of algorithms used in the con-

struction of alternative navigation strategies. While

reviewing the existing approaches, the paper proposes

a framework that combines navigation solutions to

address the realistic modeling of task-oriented mili-

tary navigation needs. This framework addresses lo-

cal and global navigation issues, the handling and

avoidance of static and dynamic obstacles, the exist-

ence of multiple military agents along with their nav-

igation needs, besides the possibility of semi-auton-

omy and user interaction in a simulation scenario.

Moreover, the paper also proposes the description of

large realistically terrains to allow the planning algo-

rithms to maintain a global vision of the scenario al-

lowing for better results in the pathfinding.

ACKNOWLEDGMENTS

We thank the Brazilian Army for the financial support

through the SIS-ASTROS Project (813782/2014), de-

veloped in the context of the PEE ASTROS 2020.

Semi-Autonomous Navigation for Virtual Tactical Simulations in the Military Domain

449

REFERENCES

Algfoor, Z. A., M. S. Sunar and H. Kolivand, 2015. A com-

prehensive study on pathfinding techniques for robotics

and video games. International Journal of Computer

Games Technology 2015: 7.

Botea, A., B. Bouzy, M. Buro, C. Bauckhage and D. Nau,

2013. Pathfinding in games. Dagstuhl Follow-Ups,

Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik.

Botea, A., M. Müller and J. Schaeffer, 2004. Near optimal

hierarchical path-finding. Journal of game develop-

ment 1(1): 7-28.

Cui, X. and H. Shi, 2011. A*-based pathfinding in modern

computer games. International Journal of Computer

Science and Network Security 11(1): 125-130.

Dooms, A., 2013. Parallel multi-agent path planning in dy-

namic.

Feurtey, F., 2000. Simulating the collision avoidance be-

havior of pedestrians. Master's Thesis.

Fiorini, P. and Z. Shiller, 1998. Motion planning in dynamic

environments using velocity obstacles. The Interna-

tional Journal of Robotics Research 17(7): 760-772.

Fletcher, J., 2009. Education and training technology in the

military. Science 323(5910): 72-75.

Frey, A., 2015. Context Steering: Behavior-Driven Steering

at the Macro Scale. Game AI Pro 2: Collected Wisdom

of Game AI Professionals. S. Rabin. Natick, MA, USA,

A. K. Peters Ltd.

Hagelbäck, J., 2012. Potential-field based navigation in

starcraft. Computational Intelligence and Games

(CIG), 2012 IEEE Conference on, IEEE.

Hagelbäck, J., 2016. Hybrid pathfinding in StarCraft. IEEE

Transactions on Computational Intelligence and AI in

games 8(4): 319-324.

Hagelbäck, J. and S. J. Johansson, 2008. The rise of poten-

tial fields in real time strategy bots. Fourth Artificial

Intelligence and Interactive Digital Entertainment Con-

ference, Stanford University.

Heinze, C., S. Goss, T. Josefsson, K. Bennett, S. Waugh, I.

Lloyd, G. Murray and J. Oldfield, 2002. Interchanging

agents and humans in military simulation. AI Magazine

23(2): 37.

Hodson, D. D. and R. R. Hill, 2014. The art and science of

live, virtual, and constructive simulation for test and

analysis. The Journal of Defense Modeling and Simu-

lation 11(2): 77-89.

Johnson, C. and A. J. Gonzalez, 2008. Automated after ac-

tion review: State-of-the-art review and trends. The

Journal of Defense Modeling and Simulation 5(2): 108-

121.

Kapadia, M. and N. I. Badler, 2013. Navigation and steer-

ing for autonomous virtual humans. Wiley Interdisci-

plinary Reviews: Cognitive Science 4(3): 263-272.

Khatib, O., 1986. Real-time obstacle avoidance for manip-

ulators and mobile robots. The international journal of

robotics research 5(1): 90-98.

Koenig, S. and M. Likhachev, 2002. D* Lite. AAAI/IAAI

15.

Likhachev, M., D. I. Ferguson, G. J. Gordon, A. Stentz and

S. Thrun, 2005. Anytime Dynamic A*: An Anytime, Re-

planning Algorithm. ICAPS.

McGaghie, W. C., S. B. Issenberg, E. R. Petrusa and R. J.

Scalese, 2010. A critical review of simulationbased

medical education research: 2003–2009. Medical edu-

cation 44(1): 50-63.

Meyer, R., T. Andre, K. Conning Chik and G. Liming,

2014. Performance Evidence Management in Live, Vir-

tual, and Constructive Training. Journal of Applied

Learning Technology 4(4).

Murphy, S. P. and T. Perera, 2002. Successes and failures

in UK/US development of simulation. Simulation Prac-

tice and Theory 9(6): 333-348.

Nilsson, N. J., 1998. Artificial intelligence: a new synthesis.

Elsevier.

Reynolds, C. W., 1999. Steering behaviors for autonomous

characters. Game developers conference.

Sahli, N. and B. Moulin, 2005. Real-world pathfinding us-

ing agent-based simulation. Agent-Directed Simulation

Symposium, part of the SpringSim.

Silveira, R., L. Fischer, J. A. S. Ferreira, E. Prestes and L.

Nedel, 2010. Path-planning for RTS games based on

potential fields. International Conference on Motion in

Games, Springer.

Silver, D., 2005. Cooperative Pathfinding. AIIDE 1: 117-

122.

Snape, J., S. J. Guy, D. Vembar, A. Lake, M. C. Lin and D.

Manocha, 2012. Reciprocal collision avoidance and

navigation for video games. Game Developers Conf.,

San Francisco.

Stentz, A., 1994. Optimal and efficient path planning for

partially-known environments. Robotics and Automa-

tion, 1994. Proceedings., 1994 IEEE International Con-

ference on, IEEE.

Stevens, J., P. Kincaid and R. Sottilare, 2015. Visual mo-

dality research in virtual and mixed reality simulation.

The Journal of Defense Modeling and Simulation

12(4): 519-537.

Tambe, M., W. L. Johnson, R. M. Jones, F. Koss, J. E.

Laird, P. S. Rosenbloom and K. Schwamb, 1995. Intel-

ligent agents for interactive simulation environments.

AI magazine 16(1): 15.

Van Den Berg, J., S. Guy, M. Lin and D. Manocha, 2011.

Reciprocal n-body collision avoidance. Robotics re-

search: 3-19.

Van den Berg, J., M. Lin and D. Manocha, 2008. Recipro-

cal velocity obstacles for real-time multi-agent naviga-

tion. Robotics and Automation, 2008. ICRA 2008.

IEEE International Conference on, IEEE.

SIMULTECH 2018 - 8th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

450