Proceedings of the 2004 EEE

International Symposium on Intelligent Control

Taipei, Taiwan, September 2-4,2004

A Behavior-based Architecture for Unmanned Aerial

Vehicles
Miaobo Dong and Zengqi Sun

Abshacr - This paper proposes an overall behavior-based con-

trol seheme for unmanned aerial vehieles (UAVs), from eoneept
.
communication with station, by which tasks are assigned,
operation are transfared and plan are customized
to implementation. Task plan, event disposal and vehicle-user Behavior-based architecture, \r3licb are nowadays success-
cooperation arc integrated in the architecture. The work is of
fully applied in robotic control""', evolved from conventional
three parts: First, basic issues for behavior-based architecture
are discussed, including delinition of behaviors, behavior attificial intelligenceas wmbmation with automation is wanted.
modes, the concept of schema and user intervention. Seeond,we In the architechre, wntrol fimction are itemized / componen-
give a brief version of software-psrt of the architecture, ma!+ tialized that is, distributed to a variety of function items, and the
out an overall family of sehema for UAV. The schema, whieh whole function is h m these items by different slructmal com-
defined BS an event-driven behavior program, is designed to binations. The operation of the agency is reactive - event-driven.
conduct the UAV operation. Third, the system strnehlre for the More about behavior-based architecturesare in [l-61.
arehitrrture is proposed consisting of ground station (users),
schema block and behavior executor. We make B major disrus-
Literature [7] gives investigation much near to this paper,
sion on behavior execution, giving a method linking UAV be- presenting a behavior-based control smcture for helicopters
haviors with flight control system that used to be complicated with background of a task contest.
and hiersrchicsl. The idea of eomponentidiazation is involved. Our works base on the following understanding: a UAV
Human-machine rooperation is a strong C O O C L I ~ and ides of works by continuously making decisions and making behaviors,
layered rooperation is employed io this paper. which are considered as segmentsof signals. The paper is on the
question: how a UAV makes decisions, and how it makes be-
haviors.
I. INTROCDUCTION Three main parts of this paper are: 1) interpretation of be-

U AVs have undergone a rapid progress in the past decades.

They are more likely to take pafl in tasks too risky for hu-
man participation and prospected to be an integral part in the
havior-based architedure with some extension issues like b e
havior defition, behavior modes, introduction of schema and
user-agency relation. 2) giving a brief version of an overall
future warfare. scheme family for UAVs with two simple tasks and two possi-
An overall control scheme for UAVs needs to implement lid ble events. A discussion on user intenention and schema r a n -
autonomous functions. These tasks used to be implemented by figmtion is c h e d oat. 3) presenting the struchre of behav-
pilots as listed below should be transferred to the on-board wn- ior-based UAV system consisting of onboard part and ground
troller. part. The mechanism of behavior executor - bow behaviors are
planning for task, automatically creating operation program executed by tlight control system, is presented. The idea of

. as a task assigned to the UAV,

event disposal, automatically execute special operations to
deal with some emergent arisen like obstacle, enemy unit,
wnpnentialization is employed.
Human-machine woperation is a strong wncem a m s s the
paper. Multiple roles buman plays are discussed The idea of
and device failure as well; layered cooperationis mainly inherited fiom [ 8 ] .

Manusmipi received May 22.2004. This work was jointly supported by 11. BEHAVIOR-BASED ARCHITECTURE
the National High Technique Ro~am(2003AA118402).the National Key
Project for Basic Research of China (G2003cb312205), the National Sci- In traditional control, control signal are generated by a con-
a c e Foundation of china(60174018 , 60305008 , 60334020 , trol law as response to the feedback from the object or the envi-
90205008 ) ronment. A single rule is bold in the whole pmcess. In behav-
Miaobo Dong is with Graduate School ai Sheuzhen. Tsinghua Univer- ior-based architecturethe signal is created in terms ofbehaviors.
sity, S h a d e n , 518057, P.R.C.(phone: +86-755-26036870: e-mail The whole control process is conjunct by different periods of
mbdon&@l63,net)
Zengqi Sun is wih State Key Laboratory of lntelligat Technology and control functions -behaviors. The agency changes its behaviors
Systems, Tsioghua University, Belling 100084, P.R.C. (phone: in the pmcess, switching between multiple control rules and is
186-10-62788939;e-mail: szq-dfr~mail.tsinghua.educn)

0-7803-8635-3/04/$20.0002004 IEEE 149

not limited in a single one. Traditional control method is gena- signal segment represents the operation SM.The behavior may
ally for one simple object or plant. The behavior-based archi- be considered as reflexive operation. User operation is regarded
teCtme is generally for an intelligent agent and a rather complex as a third kind of behavior, by which the user takes over the
environment with a series of tasks. The whole is a control execution of the behavior, and directly operates in replacement
scheme with behavior planning and event higgenhg. Robotics is of the machine agency d c h therefore transforms into a simple
the best area for its application. operation passer. It can be h o w as machine borrowing skills
from human. And it is the channel by which human and agency
A . Behavior cooperates in the opeation layer (human machine cooperation
in the architecturewill be discussed later).
Behavior can be defined h m dual sides: extend and inter-
nal, corresponding to the noun behavior and the verb be B. Schema
have. As to a behavior agency, a human or a machine agent, a
behavior is afunction between the sensing itput awl the miion A schema is a program of behavior that is used to conduct the
ourput infernally and a segment of the s i p / e x t e d i y . char- agencys operations. Regular plan aligns behaviors in series that
acteristics and skills of a behavior are contajned in its intemal occm one after another. However, as various events need to be
law and exhibited by the shape of its si& output. Changing of dealt with, which lead to diverse processing branches, series
behaviors is function switching internally and signal segment t p e plan is no more competent and pattem like schema is em-
conjunction extemally. The regular definition of behavior - ployed.
senseaction connection is mainly from the side of internal As one event occms, switching fium cnrrent behavior to an-
mechanism However, this paper will go on discussion on both other one is performed Events have extensive meanings and
sides. represent various situations. They are classified into three cate-
A behavior is generally related to a set of parameters, usually gories.
representing the state of agency, the objective and the perform- 1) environment events, those determined by environment,
ance. For example, a walking behavior of a mobile robotics may such as enemy emergence, obstacle arisen, etc, and device fail-
be expressed by walk(4bj m which a and b are current and ure included as well. They are observes in sensor input as
objective position. Additional factors like walking speed may be some exceptive changes appear. Events of this kind are those of
taken account as well according to practice. Parametem set the general sense. Handling of them is emergent and reactive.
propem of the in& function that is shown in shape of the 2) scheduled events, those that is scheduled for sake of b e
havior programming,to m g e the end of one behavior and the
beginning of another one. They define ending conditions of
behaviors, such as goal place reaches, segment operation
completed, stopped. etc.
3) user events, those generated by user intent for task as-
signment opmtion intervention or any need for mimic event.
There are two general types: code for task assignment and op-
eration request.
A schema is composed a set of behaviors and events. Each
I
behavior is associated with a number of events that may occw
Fig I Modffofkhavias during the behavior. Each event higgers a next behavior. Re-
Behaviors can be classified in three categories by principle of garding a behavior as a node and an event a line (that C O M ~ ~ ~ S
execution: 1) control law based behaviors (closed loop), 2) sig- to anotha node). a schema is actually a program p a t t m An
MI based behaviors (open loop) and 3) user o p e d o & as fig. 1 agency operates along the guidmce of the schema - behavior,
shom. control law based behaviors are executed by traditional event, second behavior, event, third behavior .._.walking
control principle. The behavior signal is generated by cntain through a path in the schema pattem.
control law as response to the sensor input. The execntor is ac- An example may help to illushate how a schema works. Con-
tually a controller and the mechanism is closed-looped.A signal sider a regular instance that a Unit moves f A to B and
based behavior is based on an experienced segment of signal, enwunters an unexpected obstacle midway. Current moving
without a stimuli-response function. L i e a reflexive action of behavior must be interrupted and the reactive avoiding behavior
human, the executor directly generates a period of signal based should he called instantly. Aftm the obstacle is avoided the
on experience as stimulated by some event. The shape of the moving behavior is resumed Such scenario may represented by

150
i n d o n of behavior as plotted in fig. 2. A schema is con- recon&uation because there are situations necessary for user to
structed to illustrate the handling as shown at the underside. It's alter schemas dpmically. Users in the architectureplay multi-
a pattern work of behaviors and events. Behaviors of shifl and ple roles of an assistant operator, a momunder and a schema
avoiding are represented hy two nodes. The occ-ce of the maker. More detailed description will be given in later discuss
event and calling to the instant avoiding operation is represented about UAVs.
by a connection fonn shifl to avoiding. Another connection back User-agency cooperation relies on communication. For
represents the resuming of moving. It is oflen a p m of a h o l e long-range systems like UAV's, the lag and other drawbacks
schema, representing a common way of event disposal. An in- need to he considered. Reasonable way of cwperation is the
tegral schema for more complicated tasks and various events is user for discrete task assignment and the agency for real-time
much more complex. operation unless some critical matter arises and user pilot has
better confidence.

In. A BEHAVIOR-BASED SCHEMA FAMILY FOR

UAV
unexpected obs12icle obsaclc Bund obasclcaroided As to a UAV for multiple tasks, an overall schema family
(behavior insertion) needs to he constructed for these tasks and various events. A
simple version of schema family for a UAV is presented here.
The UAV is projected for two tasks: 1) reconnaissance, flyingto
a specific location md rehieve useful infomatin& 2) shift,
moving iiom one base to another. Two general events are
obstacle found A-% wanted to handle: emergence of obstacle and suspicioustarget
Behaviors and schemas are Listed in table 1 each with a iden-
(SCHEMA) tifylng number. Nine behaviors are d e k e d for UAV to run at
all stages. Ibe subaerial part consists of move, accelerate and
Fig 2 Behavior insertionand &em decelerate, providing subaerial operations for taking off and
A schema may call another schema as an item in its panem. landing. The aerial part consists of IiJitdescend, hod to, hover
The propnty of schema is much like s o h a r e program, each and track. Track here includes any trajectoly-trackingoperations.
behavior is like a sentence in the program, a schema is like a Parameters for behavior are listed right side. The behavior pho-
function in the program and an event is a call to a new behavior rogrqh represents the spy operation, combined with hover as
or schema. A schema-based mechanism therefore is an hovei+photograph. The ready state of UAV is considered as an
eventdriven program. As compared with conventional hehav- enduring behavior.
ior-based reactive mechanismfM'.the latter is an event-driven Panems of all schemas are shown in fig. 3. Each is decom-
automation. posed into behaviors and sub-schemas. The main schema (17)
consists of ready, recon and ship. The UAV is initially m ready
C. User and agency state. Code oftask fiom user biggers it to execute corresponding
task, and retum to ready state afler the task is accomplished
Relationship between user and agency used to slide to hvo Schema recon and sh@, each has two results - successful exe-
opposite directions: full autonomy or teleoperation. The former cution or abortion because ofpotential danger. As danger arises,
exclude the user intervention and the latter reniove any automa- the UAV give up the task and choose to r e m at that time. Two
tion ofthe agency. possible events of environment are considered and programmed
This paper holds a view of humanmachine cooperation and in schemaJy ro, in whjch two reactive loops are designed. Ifan
mostly inherits the idea of layered cooperation fiom [8]. Users obstacle is found an avoiding wwse - schema avoid is called
and agency keep cooperation in different layers - operation and if a sospicioustarget is found, a detection (schema detect) is
layer. schema layer and beyond-schema layer. In operation layer, performed to determine whether the object is safe or dangerous,
the user sometimes takes the direct operation of the agency. resulting m different consequent handling. Two basic schemas
USWoperations are directly transferred as action outputs of the of fakeofand land are programmed as a snies of regular opera-
agency, known as mode (3) behaviors In schema layer, user? tions.
and agency interact by events. User order or request are treated
as event and triggered corresponding schema branches. A be-
yond-schema layer cooperation is defined for online schema

151
0 ready none lotakeoff height 1,2;1,6
1 move - .wsition
tareet 11 land oosition. orientation 1.3.4.7
.,_
2 accelerate acceleration I 12flyto target place 7,13,14
3 decelerate deceleration 13 avoid haghtof obstacle 4,6
4 liWdescend lift rate 14 detect target position 739
5 head to oricntatiodtaxget I5 recon target position 7,8,9,10,11,12
6 hover roll angle 16 shifl position 7,10,11,12
7 hack trajecto, 17 main none 0,15,16
8 photograph None

Fle3 Schema Famik faUAV

*que=
10 speckfir height B resched

takeoff Euspiciour
order taget
A re ched
isbed (3)
(4) I4 Evqicio" rare
La"
detect C dangcmvr
5
detect s ~ ~ p i ~ i o n s t a l g e t mtvn A

(dynamic reconfiguration)

152
 W . USER INTERVENTION AND SCHEMA RECON- V. BEHAVIOR-BASED UAV SYSTEM
FIGURATION Two points are concluded for the UAV system. 1) Operation
As stated in section users are cooperative and multi- of the aerial vehicle is conducted hy an integral schema family
fold-functional part in the architecture. Cooperation between that consists of programs for tasks. 2) Users keep layered cwp-
users and the aircrafl in three layers are kept. eration with the vehicle.
In operation layer. the user plays d e of tele-pilot. Fig. 5 shows the diagram of a behavior-based UAV system,
when the aircraft drop into some serious situation such as an separated in two parts: ground part and onboard part. The
off n o d state that need expetienced manual plot to help out. ground part (station) includes users for different duties. The
The user takes charge of pilot in replacement of the autonomous onboard part includes any device on the aircraft. A family of
system instantly. Any way, it's a way of remote control. and the schemas is installed in the controller, conducting the running of
associativeproblem of communication need to be considered. the executor. The latter translated behavior into specific control
In schema layer. the use^ iostructs the nmning progress of function, interacting with sensor and actuator. The schema block
schema by sending command codes. These codes are treated as and the executor are key components ofthe controller.
events like any other environment events, triggning schema GROUND 1 ONBOARD
entering corresponding execuiion branches.
The user and the UAV cooperate in another higher layer -
beyond-schema layer. Schemas for UAV, which was built in the
onboard controller, need to he reconligured as some exceptive
situations arise. On this occasion, the user sends reconfiguration
codes to the flying aircraft and make online alteration on the
schema, to make it fitted for the new condition, for example,
inserting a new mission, canceling a scheduled task, simplifying
handling of some events. Oynamic alteration made on the
schema, is corresponding to revising opmtion on the schema
pattan like cutting, adding an event connection, removing,
inserting a new node of behavior or subschema
An instance is presented here to illustrate how reconligma-
tions are made on an original schema for several different con- Fig 5St"eoftheUAVsynem
ditions. Fig. 4 shows a schema for task of shifl with considera- The schema block follows the event-driven principle. It d e
tion of event of suspicious target. Four possible altmtions under tects environment events in the sensor input, which represents
different conditions are desnibed on the origind pattem. an extensive collection of measured information including the
1) The user directly pilots the aircraft aAer its takeoff and re- aircraA state, the state of device and any acquisition h m the
fum autonomous landing as the goal place is reached Any environment. A part of scheduled events defined as obtainment
events that occm in the flying, is disposed by manual pilot. of specific aircraft state, are judged if activated hy the sensor
2) In repeated executions, as the environment has been input as well. User instructions are treated as events as well as
known well and potential danger excluded, the aircratl ignore dehed in schemas, higgering corresponding execution
the warning of suspicious target to simplify the program and branches.
improve the efficiency. As (2) shows, the bandhg comections Users are effective multifold-functional part in the system. As
are pruned. briefly illmated here, they are classified into commander and
3) To employ a more cautious policy, the aircraft r e m base pilot for different duties. ?be commander interacts with the
without finther detection as any suspicious target arises. Acon- schema block, sending insauctions to the block to control the
nection fnmfly to to refumis added for the event of suspicious nmning progress, and sending reconfigurationcodes to alter the
target. schema if necessary. The pilot take remote manipulation of the
4) The user assigns an additional task - to detect a specific executor as manual operation is needed. when the commander
location in the mid-way. An execution branch is added triggered decides to insert a tele-pilot session, he sends an operation re-
hythe user order of assignment. quest to the schema block to pause the schema processing and
The pattern-like program provides an convenient way for sends a command to the pilot at the same time. So the pilot can
schema rewniiguration. take charge ofthe executor in replacement ofthe schema block.
when the pilot is end a notification is sent w resume the
schema block.

153
 A C n V A ' l l n N l 1RT

I ITl3lFAMILY

The executor is a module that implements behaviors. A item).

method of translation must he developed fiom behavior to wn- The total control function is yielded by combination of these
trol action. Flight control system contains various control func- items. h overall configuration contains o d o f f state of every
tions Like pitch control, roll control, altiMe control,etc. They item and parameter for it. Different combination of activated
are hierarchically integrated in much complex@'. A behavior items and their parameter yield diEerent integral control func-
execution should make arrangement and sening of these control tion. Each behavior may he translated into a specific c o n f i p -
funcfions. We use a componential designto solve this problem. tion ofitems.
Fig. 6 illusbates the mechanism of the executor. Flight wn- As a behavior is inpuited the translator maps it to an activa-
trol complexity is componentialized into items. Each of these tion list for all these items, deciding a s e l d o n of activated
items implements a simple control functions. Basic control items and thkr parameters. When the list is applied to the item
functions and upper-level control function as well, are capsu- family, the behavior is executed.
lated in these items listed as A l , A2, B1, ... Every item is re- As the user iutmenes, he may manipulator the executor hy
garded as a component with specific interface. ?he stNchue of seniug the activation list to customize the control function or
the system is represented by their collaborative connection. directly settingthe output signal to actuators.
For every item a set ofparameters is related to it to configme
its function. The parameter is treated as equal to other in- VI. SUMMARY
putloutput information like sensor input and control signal (ac- This paper proposed an overall behavior-based architecture
tuator). Every item functions through connection to sensor input, for UAVs. The concept of schema is introduced as behavior
actuator signal and parameters. The figure describes the struc- program driven by events. Scheduling, event disposal and user
ture of the B1 item (pitch control) that employs a pmportiOnal instruction are all integrated in the schema by principle of event
control law with rate damping. The parameter for it is its objec- higgaing. To apply the schema-based mechanism. we conshllct
tive s e h g . Other items have similar principles. a simplified family of schemas to conduct the operation of a
These items are organized as a cluster along lines of informa- UAV, and present the structure of the behavior-based UAV
tion sensor, actuator and parameter, standing for collection of
~
system.
measurement, achmtors signals, and parameter for all items
respectively. Basic control items Like BI, B2 have directly out-
put signal to actuators. Upper-level items like Cl, C2 function
without connection to actuators, hut with parameter (for other

154
tematiou Coiifcroiico of I(cibotics anti Autoniation. p p 271 1-2718,
Nice, France, May 1992.
R. A. Brooks. "A Robust Layered Contml System Cor Mobile Ilohot,"
IUUL .loun,al of Robotics aid Aiitoiiiatioii, Vol. IlA-2, i>p 14-23,
1986.

155