Autonomous Mission Management for Forest Search with Multiple

Unmanned Aerial Vehicles

Kemao Peng

, Feng Lin

and Ben M. Chen

Temasek Laboratories, National University of Singapore, 117411, Singapore, Singapore

Department of Electrical and Computer Engineering, National University of Singapore, 117576, Singapore, Singapore

Keywords:

Unmanned Aerial Systems, Autonomy of UAVs, Multiple UAVs.

Abstract:

An autonomous mission management (AMM) system is designed with the enhanced hierarchical-distributed

methodology (HDM) for multiple unmanned aerial vehicles (UAVs) to search a ﬁeld of forest together. The

main ideas of the enhanced HDM are hierarchical control and distributed implementation. The event control

law is partitioned into the group and individual event control laws. The group event control law is to coordinate

the group of UAVs to complete the designated mission and the individual event control laws are to complete

the assigned submissions/ tasks accordingly. The group event control law is executed by the leader and any

member can be designated or selected as the leader on the rules. The forest search is applied to verify the

designed AMM system in simulation. The simulation results demonstrate that the designed AMM system is

successful to complete the designated mission by collaborating the group of UAVs.

1 INTRODUCTION

The autonomous mission management (AMM) has

been attracting much attention for the multiple un-

manned aerial vehicles (UAVs) as it is necessary to

coordinate a group of UAVs to complete a designated

mission together (Bellingham et al., 2002), (Inalhan

et al., 2002), (Peng et al., 2014), (Richards et al.,

2002). Such the AMM processes in the high level and

the dynamic processes in the low level of the multi-

ple UAVs construct a typical hybrid system in which

the dynamic processes are time-driven and the AMM

processes are event-driven (Kaminer et al., 2006),

(Tomlin et al., 2000), (Uhrmann and Schulte, 2011),

(van der Schaft and Schumacher, 1998). The event

variables are deﬁned to describe the system behaviors

that are extracted from the time-driven process. The

transition of the discrete event states will become part

of the event control laws to be designed. Thus, too

many deﬁned event variables may result in that the

designed event control laws are not implementable.

Many researchers dedicated their effort to the hy-

brid systems (Jadbabaic et al., 2003), (Kopeikin et al.,

2013), (Lafferriere et al., 2005), (Murata, 1989), (Teo

et al., 2004), (Wong-Toi, 1997), (Ye et al., 1998). The

typical tools such as Petri net and temporal logic are

popularly used in the logic control design of the hy-

brid systems. However, the core problems are still

left to study. One problem is how to describe the sys-

tem behaviors with reasonable number of the event

variables so that the designed event control laws are

implementable. Another problem is the conversion

of the event commands to dynamic commands as the

UAVs are capable only of tracking the dynamic com-

mands. Such conversion is subject to the capabilities

of the UAVs. Otherwise, the event commands are not

executable.

Our objective is to design an AMM system to

collaborate a group of UAVs (quadrotors) to search

a ﬁeld of forest together. Such system is described

in the hybrid systems. We try to attenuate the ef-

fect of the two problems by reducing number of the

event variables to be deﬁned and deﬁning available

event commands based on the physical process. This

motivated us to propose the enhanced hierarchical -

distributed methodology (HDM).

HDM was proposed for formation ﬂight in (Peng

et al., 2014) . In this paper, it is enhanced for more

complex scenarios so-called the enhanced HDM in

which a submission/ task is decomposed into a series

and/or parallel of tasks/ subtasks respectively as the

UAVs can execute various kinds of tasks/ subtasks si-

multaneously. Such enhancement results in that the

missions can be decomposed perfectly with the se-

ries and/ or parallel relationships in multiple levels.

The main ideas of the enhanced HDM are hierarchi-

Peng, K., Lin, F. and Chen, B.

Autonomous Mission Management for Forest Search with Multiple Unmanned Aerial Vehicles.

DOI: 10.5220/0005949603850392

In Proceedings of the 13th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2016) - Volume 1, pages 385-392

ISBN: 978-989-758-198-4

385

❄

A designated mission

Group event

control law

✲

❄

✛

❄

···· · ····

Individual event

control law

event-driven

part

✛

❄

Conversion of

event cmd to

dynamic cmd

Dynamic

control law

✛

✲

❄

✻

Time-driven

part

Control law Plant

UAV 1

Individual event

control law

event-driven

part

✛

❄

Conversion of

event cmd to

dynamic cmd

Dynamic

control law

✛

✲

❄

✻

Time-driven

part

Control law Plant

UAV n

Figure 1: A closed-loop system of multiple UAVs.

cal control and distributed implementation. The event

control law is divided into the group and individual

event control laws. The group event control law is

to focus on the group events to collaborate the group

of UAVs to complete the designated mission together.

The individual event control laws are to focus on the

individual events to complete the assigned submis-

sions/ tasks accordingly. The group event control law

is conducted by the leader. Any group member can be

designated or selected as the leader on the rules.

2 ENHANCED HDM

The enhanced HDM is introduced. The resulting

closed-loop system is shown in Figure 1. A group of

UAVs collaborate to complete a designated mission

together. The event control law consists of the group

and individual event control laws hierarchically. The

group event control law is conducted by any member

when it is designated or selected as the leader. We

mainly consider the group and individual event con-

trol laws, conversion of the event commands to dy-

namic commands and the event-driven transition.

2.1 Group Event Control Design

The group event control law is focused on the group

events such as the mission decomposition, schedule,

assignment and progress. When an accident happens,

the mission reschedule is to be considered.

2.1.1 Mission Decomposition

A mission may be decomposed into a series and/or

parallel of submissions so that it is easy to be com-

pleted by a group of the UAVs together. The mission

is completed when those submissions are completed

accordingly.

:= {s

mn,11

,··· , s

mn,ij

,··· ,s

mn,m

}, (1)

where m

denotes the mission decomposition and

mn,ij

denotes the i-th and j-th submission. There are

× n

of submissions.

2.1.2 Mission Schedule

A decomposed mission can be scheduled according

to the series and/or parallel relationships between/

among the submissions so that the submissions can

be assigned accordingly.

sch



mn,1

··· s

mn,n



mn,k



mn,1k

··· s

mn,m



′

k ∈ {1,··· ,n

(2)

where m

sch

denotes the mission schedule. The sub-

missions are assigned in n

of steps and m

of the sub-

missions are assigned in the k-th step based on the

event activation conditions.

In the presentation, the row vectors mean series

and the column vectors mean parallel. A series of the

submissions are completed one by one and a parallel

of the submissions in a step are completed in the step.

A zero element of the vectors means no correspond-

ing assignment. No assignment and no activation are

regarded as completed. The rules are applicable to

presentation of the tasks and subtasks hereafter.

2.1.3 Mission Assignment and Progress

The mission assignment and progress can be recorded

in sequence of the mission schedule so that the com-

pleted, being completed and to be completed submis-

sions are presented clearly.



2 ··· 2 1 0 ··· 0



sap



2 ··· 2 1 0 ··· 0



′

(3)

where m

denotes the step progress of the mission

and m

sap

denotes the progress of the submissions in

the being completed step. The element 2 means that

the corresponding step/ submission is completed; 1

means that the corresponding step/ submission is be-

ing completed; 0 means that the corresponding step/

submission is to be completed. The next step/ sub-

missions to be assigned is/are clear based on the event

activation conditions. The meanings of the elements

are applicable to the submission/ task assignment and

progress hereafter.

2.2 Individual Event Control Design

The individual event control laws are focused on the

individual events such as the submission/ task decom-

position, schedule, assignment and progress. when an

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

386

accident happens, the submission/ task reschedule is

to be considered.

2.2.1 Submission/Task Decomposition

A submission/ task may be decomposed into a series

and/or parallel of tasks/ subtasks respectively so that

it is easy to be completed by a UAV. The submis-

sion/ task is completed respectively when the tasks/

subtasks are completed accordingly. The subtasks are

deﬁned as those easily completed by the UAVs.

:= {t

sk,11

,··· ,t

sk,ij

,··· ,t

sk,m

:= {s

tk,11

,··· ,s

tk,ij

,··· ,s

tk,m

(4)

where s

denotes the submission decomposition and

sk,ij

denotes the i-th and j-th task. There are m

× n

of tasks. t

denotes the task decomposition and s

tk,ij

denotes the i-th and j-th subtask. There are m

× n

of subtasks.

2.2.2 Submission/Task Schedule

A decomposed submission/ task can be scheduled ac-

cording to the series and/or parallel relationships be-

tween/ among the tasks/ subtasks so that the tasks/

subtasks can be assigned accordingly.

sch



sk,1

··· t

sk,n



sk,k



sk,1k

··· t

sk,m



′

k ∈ {1,··· ,n

sch



tk,1

··· s

tk,n



tk,j



tk,1j

··· s

tk,m



′

j ∈ {1, · · · ,n

(5)

where s

sch

denotes the submission schedule. The

tasks are assigned in n

of steps and m

of the tasks

are assigned in the k-th step based on the event acti-

vation conditions. t

sch

denotes the task schedule. The

subtasks are assigned in n

of steps and m

of the sub-

tasks are assigned in the j-th step based on the event

activation conditions.

2.2.3 Submission/Task Assignment and Progress

The submission/ task assignment and progress can be

recorded in sequence of the submission/ task sched-

ule so that the completed, being completed and to be

completed tasks/ subtasks are presented clearly.



2 ··· 2 1 0 ··· 0



tap



2 ··· 2 1 0 ··· 0



′



2 ··· 2 1 0 · · · 0



sap



2 ··· 2 1 0 · · · 0



′

(6)

where s

denotes the step progress of the submission

and s

tap

denotes the progress of the tasks in the being

completed step. t

denotes the step progress of the

task and t

tap

denotes the progress of the subtasks in

the being completed step.

2.3 Conversion of Event Commands to

Dynamic Commands

The output of the event control laws is to assign sub-

tasks. Such subtasks, the event commands, need to

be converted properly so that they can be easily con-

ducted by the UAVs. There are three kinds of sub-

tasks such as (1) message subtask: It needs to deﬁne

a set of knowledge for the communication between/

among the UAVs. Once a UAV receives a message,

it can understand the message and know how to re-

spond the message; (2) action subtask: It needs to de-

ﬁne a number of actions that the UAVs can easily ex-

ecute. When a UAV is assigned an action subtask, it

can know what kind of action it needs to conduct; and

(3) ﬂight subtask: When a UAV is assigned a ﬂight

subtask, it should know how to complete the ﬂight

subtask. Thus, there needs an online planning to con-

vert the event command, a ﬂight subtask, into the dy-

namic command, the trackable references to the dy-

namic control laws. Based on such conversion, the

event command can be executed by the dynamic con-

trol laws and thus the UAVs can ﬂy on schedule.

2.4 Event-driven Transition

The event state transition is activated by the states of

the deﬁned events. Most of the events are extracted

from the time-driven process to describe the system

behaviors and thus their states depend on the time-

driven process. Therefore, it needs to deﬁne the ﬂight

subtask completion conditions so that the states of the

corresponding events can transit correctly.

Based on the ﬂight subtask completion conditions,

the states of the deﬁned events are clear. Therefore,

the activation conditions of the deﬁned events each

can be established according to the event states to

drive the event state transition.

The four parts construct the AMM system in the

high level. This completes introduction of the en-

hanced HDM.

3 DESIGN OF AMM SYSTEM

An AMM system is designed by following the ideas

of the enhanced HDM to collaborate nine UAVs

(quadrotors) to search a ﬁeld of forest with size of

500 × 500 meters together. The design is subject to

the capabilities of the UAVs and the environment sit-

uation. Next, we proceed to design the AMM system.

Autonomous Mission Management for Forest Search with Multiple Unmanned Aerial Vehicles

387

500 m

✲

✻

Y(E)

X(N)

Accessible side

✻

✲

❄

✻

✲

❄

✻

✲

❄

✻

✲

❄

✻

✲

❄

✻

✲

❄

✲ ✛

❄

✻

10 m

✲

✻

Y(E)

X(N)

x x x

1 2

xxx

Take-off and landing locations

Figure 2: The forest ﬁeld.

3.1 Group Event Control Design

The ﬁeld of forest is assumed to originate normally.

The North-East-Down (NED) frame is deﬁned in Fig-

ure 2. The x denotes the take-off/ landing locations.

The distance between them and the origin is 5 meters.

Only one side of the forest ﬁeld is accessible.

The ﬁeld of forest is partitioned into a set of ﬂight

channels in Figure 2 so that the UAVs can search

the forest by ﬂying along the channels. The nomi-

nal distance between the two neighbor channels is set

to L = 40 meters for slight overlap.

The nine UAVs are assigned in three batches and

there are three UAVs in a batch. The ﬁrst two

batches are pre-scheduled and the third batch is on-

line scheduled based on the search results of the ﬁrst

two batches. A batch of UAVs are assigned to ﬂy into

the ﬁeld of forest together. Each UAV is allocated a

round channel to search the forest independently and

records the detected area in 2D map, the detected tar-

gets in the coordinates. All of the members in a batch

meet around the launch site to hand over their data of

the established 2D maps and the found targets to the

batch leader before they land.

The batch leader is designated or selected on the

rules. The batch leader is designated at the beginning

of each batch. The ﬁrst member in the ﬁrst batch is

designated as the leader by the ground station. The

ﬁrst member in the other batches is designated as the

leaders by the leader of the last batch. There is no

batch leader during independentlysearching. The ﬁrst

UAV that arrives at its meeting point becomes the new

batch leader. A UAV is regarded lost if it cannot arrive

at its meeting point in the given duration. The group

resources are shared by all of members in a batch .

The batch leaders have six duties such as (1) call

all of the members in the batch to take off and to

search independently; (2) collect and merge those 2D

maps and the found targets; (3) schedule the path on-

line based on the merged 2D map and decide number

of UAVs to be assigned in the third batch; (4) hand

over the merged 2D map, found targets, online sched-

uled path, and assignment of the UAVs in the third

batch to the standby UAVs of the next batch; (5) des-

ignate the ﬁrst member of the next batch as the leader

before it lands; and (6) hand over the merged 2D map

and the found targets to the ground station at end of

the search before it lands.

There is no communication between the UAVs

and the groundstation during the process except at the

beginning and end of the process. At the beginning,

the ground station designates the ﬁrst member in the

ﬁrst batch as the batch leader. At the end, the ground

station receives and displays the search results of the

merged 2D map and the found targets. The UAVs can

communicate with each other if they are close. This

is the scenario of the group of UAVs.

3.1.1 Mission Decomposition

Based on the scenario of the group of UAVs, the mis-

sion is decomposed as follows,

= {s

mn,11

,··· ,s

mn,ij

,··· , s

mn,33

}, (7)

where s

mn,ij

denotes the submission to be assigned to

the i-th UAV in the j-th batch. Those submissions are

to be deﬁned latter.

3.1.2 Mission Schedule

Based on the mission decomposition, m

, the decom-

posed submissions are scheduled as follows,

sch



atch,1

atch,2

atch,30



atch,30



atch,3

atch,



′

atch,

= 0, no assignment,

atch,k



mn,1k

mn,2k

mn,3k



′

k ∈ {1,2,3} .

(8)

Based on the mission schedule, m

sch

, nine of the sub-

missions are assigned in three batches and three of

them are assigned together in a batch based on the

event activation conditions.

The mission assignment and progress vectors

record the completed, being completed and to be com-

pleted batch/ submissions. The next batch/ submis-

sions to be assigned is/are clear based on the event

activation conditions. The group event state transition

is shown in Figure 3. The series of the batches are

are completed one by one. The third batch is acti-

vated based on the online decision. A parallel of the

submissions in a batch are completed in the batch ex-

cept that the assigned UAV is lost.

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

388

Start

Batch

Stop

✲ ✲ ✲ ✲

✻

Figure 3: Group event state transition.

3.2 Individual Event Control Design

Each UAV is assigned a submission. The submissions

are deﬁned such as (1) receive the related data and

stay on the ground; (2) take off and independently

search the forest along the scheduled channel/path;

(3) broadcast itself ﬂight motion messages; (4) record

the detected areas in 2D map and the found targets

in the coordinates; (5) ﬂy to and hover at the desig-

nated meeting points and attend the selection of the

new batch leader; (6) hand over the recorded data to

the batch leader; and (7) land on the ground and exit.

The submissions of the UAVs in the third batch

are slightly revised only at the ﬂight path that is online

scheduled. The submissions of the UAVs as the batch

leader are also revised. Those UAVs do not need to

hand over the recorded data to the batch leader. But

they have to undertake the duties of the batch leader.

This is the scenario of each UAV.

3.2.1 Submission/ Task Decomposition

Based on the scenario of each UAV, the submissions

are decomposed as follows,

mn,ij

= {t

sk,11

sk,21

sk,12

sk,22

sk,32

sk,13

sk,23

sk,4

sk,15

sk,25

}, i, j ∈ {1,2, 3},

(9)

where the decomposed tasks are listed in Table 1.

Note that s

denotes submission, t

denotes task and

denotes subtask.

Table 1: Decomposed tasks.

Task No. Tasks

11 Stay on the geound

21 Rv the related data

12 Take off/ search

22 Tr the motion messages

32 Record the 2D map/targets

13 Hv at the meeting point

23 Hand over the data

4 Landing

15 Landed on the ground

25 Report the landed

Rv receive

Tr transimit

Hv hovering

Table 2: Decomposed subtasks in Task 21.

Subtask No, Subtasks

1 Rv path/Rp

2 Rv agn/Rp

3 Rv 2D map/Rp

4 Rv tgt data/Rp

5 Rv leadership/Rp

6 Tr take-off cmd/Rv Rp

7 Tr group data

8 Rv take-off cmd/Rp

9 Rv group data

Rp respond/resoinse

cmd command

agn assignment

tgt target

The tasks may further be decomposed as follows,

sk,21

= {s

tk1,1

,··· ,s

tk1,9

sk,12

= {s

tk2,1

,··· ,s

tk2,n

st2

sk,23

= {s

tk3,1

,··· ,s

tk3,17

sk,4

= {s

tk4,1

tk4,2

(10)

where n

st2

= 7 or to be determined online. Task 12

consists of n

st2

line segments at which each segment is

as a ﬂight subtask.Task 4 consists of 2 line segments.

Task 21 and Task 23 consist of the message subtasks

which are listed in Table 2 and 3 respectively. The

other tasks are as simple as the subtasks and thus no

decomposition is needed.

3.2.2 Submission/ Task Schedule

Based on the submission decomposition, s

mn,ij

, the

decomposed tasks are scheduled as follows,

sch,ij



sk,1

sk,2

sk,3

sk,4

sk,5



sk,1



sk,11

sk,21



′

, i, j ∈ {1,2,3},

sk,2



sk,12

sk,22

sk,23



′

sk,3



sk,13

sk,23



′

sk,5



sk,15

sk,25



′

(11)

Based on the task decomposition, t

sk,ij

, the decom-

posed tasks are scheduled as follows,

sch,21



tk1,10

tk1,50



tk1,10



tk1,11

tk1,21

tk1,31



′

tk1,11

= 0, no assignment,

tk1,21



tk1,3

tk1,4



tk1,31



tk1,1

tk1,2

tk1,3

tk1,4



(12)

tk1,50



tk1,15

tk1,25



′

tk1,15



tk1,5

tk1,6

tk1,7



tk1,25



tk1,8

tk1,9



Autonomous Mission Management for Forest Search with Multiple Unmanned Aerial Vehicles

389

Table 3: Decomposed subtasks in Task 23.

Subtask No. Subtasks

1 Iq leadership

2 No Rp to Iq

3 Rp Iq/Rv Rp

4 Rq 2D map/Rv

5 Rq tgt data/Rv

6 Tr lnd cmd/ Rv Rp

7 Merge map/tgt data

8 Online scheduling

9 Tr path/Rv Rp

10 Tr agn/Rv Rp

11 Tr 2D map/Rv Rp

12 Tr tgt data/Rv Rp

13 Tr leadership/Rv Rp

14 Rv Rp to Iq/Rp

15 Rv Rq/Tr 2D map

16 Rv Rq/Tr tgt data

17 Rv lnd cmd/Rp

lnd landing

Iq inquire/inquiry

Rq request

sch,12



tk2,1

··· s

tk2,n

st2



sch,4



tk4,1

tk4,2



(13)

sch,23



tk3,1

tk3,20



tk3,20



tk3,120

tk3,220



′

tk3,120



tk3,14

tk3,15

tk3,16

tk3,17



tk3,220



tk3,2

tk3,30

tk3,7

tk3,80

tk3,110



tk3,30



tk3,130

tk3,230



′

tk3,80



tk3,18

tk3,28



′

tk3,110



tk3,111

tk3,211



′

tk3,18

= 0, no assignment,

tk3,28



tk3,8

tk3,9

tk3,10



tk3,k30



tk3,3

tk3,4

tk3,5

tk3,6



tk3,k11



tk3,11

tk3,12

tk3,13



k ∈ {1,2}.

(14)

Based on the submission/ task schedules, the decom-

posed tasks/ subtasks can be assigned accordingly.

The submission/ task assignment and progress vectors

record the completed, being completed and to be com-

pleted step/ task/ subtask. The next step/ task/ subtask

is clear based on the event activation conditions.

The submission event state transition is shown in

Figure 4. The event state transition in Task 21 is

shown in Figure 5. There are three branches from the

enter for the three batches respectively. The UAVs in

the ﬁrst batch do not need to receive the online sched-

uled path and assignment, the merged 2D map and

targets. The UAVs in the second batch do not need

Enter

ask

End

✲ ✲ ✲ ✲ ✲ ✲

Figure 4: Submission event state transition.

to receive the online scheduled path and assignment,

but they need to receive the merged 2D map and tar-

gets. The UAVs in the third batch if applicable need

to receive the online scheduled path and assignment,

as well as the merged 2D map and targets. The subse-

quent two branches are for the batch leader and mem-

bers respectively.

The event state transition in Task 23 is shown in

Figure 6. The UAVs in a batch attend the leader selec-

tion for the new batch leader. There are two branches

of the outcome. One is for the members and the other

is for the new batch leader. The leader needs to col-

lect the data from the two members respectively in

two branches. After merging the 2D map and found

targets, the leader conducts the online schedule only

in the second batch. Thus, one branch is no assign-

ment and the other is for the online schedule. Sub-

sequently, the leader needs to hand over the merged

map and found targets to the standby UAVs in the next

batch in one branch and in the other branch hand over

them to the ground station at end of the search. This

completes the individual event control design.

Enter

ubtask

1, 1

ubtask

1, 2

ubtask

1, 3

ubtask

1, 4

✲ ✲ ✲ ✲

❄

❄❄

ubtask

1, 5

ubtask

1, 6

ubtask

1, 7

ubtask

1, 8

ubtask

1, 9

End

✛✛✛✛

✛✛

✻

Figure 5: The event state transition in Task 21.

Enter

ubtask

3, 1

ubtask

3, 14

ubtask

3, 15

ubtask

3, 16

ubtask

3, 17

End

✲ ✲ ✲ ✲ ✲ ✲

ubtask

3, 2

ubtask

3, 3

ubtask

3, 4

ubtask

3, 5

ubtask

3, 6

❄

✲ ✲ ✲ ✲

ubtask

3, 3

ubtask

3, 4

ubtask

3, 5

ubtask

3, 6

✲ ✲ ✲ ✲

ubtask

3, 7

ubtask

3, 8

ubtask

3, 9

ubtask

3, 10

❄❄

✛✛✛✛

❄❄

ubtask

3, 11

ubtask

3, 12

ubtask

3, 13

✲ ✲ ✲ ✲

ubtask

3, 11

ubtask

3, 12

ubtask

3, 13

✲ ✲ ✲

✲

✻

Figure 6: The event state transition in Task 23.

3.3 Conversion of Event Commands to

Dynamic Commands

Eighteen sets of the knowledge are deﬁned for the

message subtasks and two sets of the action are de-

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

390

ﬁned for the action subtasks. The ﬂight subtasks are

presented as the line segnebts in the starting and end

points and thus they need to be converted into the

trackable references of the position, velocity and ac-

celeration by the UAVs.

There may be four phases to complete the ﬂight

subtasks such as the acceleration (Ac), velocity hold-

ing (Hd), deceleration (Dc) and hovering (Hv). The

ﬁrst three phases are velocity tracking and the last

phase is position tracking.











Hv, d ≤ d

Ac+ Hv, d

< d ≤ 2d

Ac+ Dc+ Hv, 2d

< d ≤ d

Ac+ Hd+ Dc+ Hv, d > d

(15)

where d

denotes the distance that the UAVs can

hover from one point to another point. d

denotes

the distance that the UAVs need to hold the maximal

speed to ﬂy. d

= v

max

with v

max

and a

max

be-

ing the maximal speed and acceleration. d denotes the

distance between the starting point, p

, and end point,

. d = kp

− p

k. The trackable references can be

computed well for the four phases each. The heading

reference is the direction of the ﬂight subtask from the

starting point to the end point in the ﬁrst three phases

and it is the direction of the next ﬂight subtask in the

last phase or pointing to the North if the next ﬂight

subtask is hovering. With such conversions, the event

commands are executable.

3.4 Event-driven Transition

A set of the event activation conditions are deﬁned

to control the transition of the discrete event states.

The ﬂight subtask completion conditions are deﬁned

based on the distance along the direction of the ﬂight

subtask. The ﬂight subtask is completed when

≤ 0, d

= (p

− p)

′

− p

)/d,

where p denotes the position of the UAV. d

denotes

the projection of the distance to be ﬂied relative to the

end point to the direction of th ﬂight subtask. If the

ﬂight subtask is hovering, it is completed when the

hovering is over the given duration. With the ﬂight

subtask completion conditions, the transition of the

discrete event states is clear.

4 SIMULATION

The simulation is conducted to verify the designed

AMM system to coordinate nine of the UAVs to

search the ﬁeld of forest. The resulting closed-loop

system is shown in Figure 7 in which Gds denotes

Gds Cmm Env

UAV 1 UAV 2 UAV 3 UAV 4 UAV 5

UAV 6 UAV 7 UAV 8 UAV 9

Figure 7: A simulation system of multiple UAVs.

the ground station, Cmm denotes the communication

system and Env denotes the surroundingenvironment.

In the simulation,, one UAV is assumed lost in

the ﬁrst batch and another UAV is lost in the second

batch. Based on the merged map, the online schedule

decides to assign two UAVs and schedules the path

for them each to search the missed areas in the third

batch. Then, the full of the forest ﬁeld is searched.

The discrete event states and ﬂight trajectories of

the UAVs are shown in Figure 8. The 2D maps built

and merged by the UAVs to describe the detected ar-

eas are shown in Figure 9. The simulation results

demonstrate that the designed AMM system is suc-

cessful to coordinate the group of UAVs to search the

ﬁeld of forest together. The number of the deﬁned

events is not too many to affect the implementation of

the AMM system. The designed AMM system is also

successfully veriﬁed in our high-ﬁdelity simulator.

5 CONCLUDING REMARKS

The enhanced HDM has been successfully applied to

design an AMM system to collaborate multiple UAVs

to complete a designated mission together. The main

features of the designed AMM system are hierarchi-

cal control, series and/or parallel decomposition and

distributed implementation. The missions can be de-

composed perfectly with the series and/or parallel re-

lationships in multiple levels. The enhanced HDM

is applicable to the other more complex scenarios.

Nonetheless, the mission reschedule is to be studied

when an accident happens.

REFERENCES

Bellingham, J., Tillerson, M., Alighanbari, M., and How, J.

(2002). Cooperative path planning for multiple uavs in

dynamic and uncertain environments. In Proceedings

of the 41st IEEE Conference on Decision and Control,

pages 2816–2822, Las Vegas, Nevada, USA. IEEE.

Inalhan, G., Stipanovic, D., and Tomlin, C. (2002). De-

centralized optimization with application to multi-

ple aircraft coordination. In Proceedings of the 41st

IEEE InternationalConference on Decision and Con-

Autonomous Mission Management for Forest Search with Multiple Unmanned Aerial Vehicles

391

trol, pages 1147–1155, Las Vegas, Venada, USA.

IEEE.

Jadbabaic, A., Lin, J., and Morse, A. S. (2003). Coordi-

nation of groups of mobile autonomous agents with

neighbor rules. IEEE Transactions on Automatic Con-

trol, 48(6):998–1001.

Kaminer, I., Yakimenko, O., Pascoal, A., and Ghabcheloo,

R. (2006). Path generation, path following and co-

ordinated control for timecritical missions of multiple

uavs. In Proceedings of 2006 American Control Con-

ference, Minneapolis, MN, USA. IEEE.

Kopeikin, A. N., Ponda, S. S., Johnson, L. B., and How,

J. P. (2013). Dynamic mission planning for commu-

nication control in multiple unmanned aircraft teams.

Unmanned Systems, 1(1):41–58.

Lafferriere, G., Williams, A., Caughman, J., and Veerman,

J. E. (2005). Decentralized control of vehicle forma-

tions. Systems and Control Letters, 54(9):899–910.

Murata, T. (1989). Petri nets: properties, analysis and ap-

plications. Proceedings of the IEEE, 77(4):541–580.

Peng, K., Pang, T., Lin, F., and Chen, B. M. (2014). Au-

tonomous mission execution for multiple unmanned

aerial vehicles with hierarchical-distributed method-

ology. In Proceedings of 2014 11th IEEE Interna-

tionalConference on Control and Automation, pages

1369–1374, Taichung, Taiwan. IEEE.

Richards, A., abd M. Tillerson, J. B., and How, J. (2002).

Coordination and control of multuple uavs. Aiaapaper

2002–4588.

Teo, R., Jang, J. S., and Tomlin, C. J. (2004). Automated

multiple uav ﬂight the stanford dragonﬂy uav pro-

gram. In Proceedings of the 43rd IEEE Internation-

alConference on Decision and Control, pages 4268–

4273, Atlantis, Geogia, USA. IEEE.

Tomlin, C. J., Lygeros, J., and Sastry, S. S. (2000). A game

theoretic approach to controller design for hybrid sys-

tems. Proceedings of the IEEE, 88(7):949–969.

Uhrmann, J. and Schulte, A. (2011). Task-based guid-

ance of multiple uav using cognitive automation.

In Proceedings of 2011 International Conference on

Advanced Cognitive Technologies and Applications,

Rome, Utaly. ABIA.

van der Schaft, A. J. and Schumacher, H. (1998). Comple-

mentarity modeling of hybrid systems. IEEE Trans-

actions on Automatic Control, 43(4):483–490.

Wong-Toi, H. (1997). The synthesis of controllers for lin-

ear hybrid automata. In Proceedings of the 1997 IEEE

Conference on Decision and Control, pages 4607–

4613, San Diego, CA, USA. IEEE.

Ye, H., Michel, A., and Hou, L. (1998). Stability theory

for hybrid dynamical systems. IEEE Transactions on

Automatic Control, 43(4):461–474.

APPENDIX

−200 −100 0 100 200

100

200

300

400

500

(m, North)

(m, East)

0 500 1000 1500 2000 2500

Event states

Time (s)

−200 −100 0 100 200

100

200

300

400

500

(m, North)

(m, East)

Batch 1

Batch 2

Batch 3

Number of the defuned event states is 102.

13 of the event states are used for each UAV.

2 of the event states are for the ground station.

One UAV is lost in Batch 1.

Another UAV is lost in Batch2.

Batch 3 is scheduled online based on

the search results of the first two batches.

Batch 2 Batch 1

Figure 8: Event states and ﬂight trajectories of UAVs.

200 250 300

100

(m, East)

(m, North)

0 200 400

100

200

300

400

500

(m, North)

0 200 400

100

200

300

400

500

(m, North)

0 200 400

100

200

300

400

500

0 200 400

100

200

300

400

500

(m, East)

0 200 400

100

200

300

400

500

0 200 400

100

200

300

400

500

0 200 400

100

200

300

400

500

(m, East)

(m, North)

UAV(1,1)

UAV(2,1)

UAV(1,2)

UAV(3,2)

UAV(1,3)

UAV(2,3)

Gds

Sensor detected area

1: detected

0: undetected

1st row: merged maps.

2nd row: maps of single UAV.

1st batch:UAV(1,1) and (2,1).

2nd batch: UAV(1,2) and (3,2).

3rd batch: UAV (1,3) and (2,3).

Gds: ground station.

10 targets are detected.

Targets

Figure 9: Maps of the detected areas by UAVs.

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

392