07-01-2012, 03:04 PM
HYBRID BEHAVIOURAL ARCHITECTURE
The behavioural approach is based on the existence of
individual behaviours and on the coordination of
these behaviours. It states: autonomy emerges from
the co-operative work of various behaviours [l].
We define a behaviour as an independent stereotyped
action that is maintained by a specific perceived
stimulus. Examples of robot navigation behaviours are
Wonder arortnd or Go along. Each behaviour is activated
by a stimulus. The overall agent behaviour emerges
from the coordination of the various active behaviours.
The behavioural approach is of special interest for
building agents which interact strongly with their
environment by sensing.
Figure 2 illustrates the basics of a the behavioural
architecture which derives from this approach. We
recognise a structure organised in three levels of
abstraction:
Level 1: physical
Level 2: behavioural
Level 3: cognitive
Level 1 includes all devices for acting on the robot and
for sensing the environment.
Level 2 is the heart of the architecture and implements
the behaviours as independent modules.
Level 3 includes units which implement the tasks to be
performed by the agent. The units proceed by
activation/deactivation of the behaviours. Units can be
built according to very different principles [6].
In the frame of this paper, we select the principle of a
task described by a state machine whose transitions
are controlled by a status vector and each state gives
rise to an activation vector. Status and activation
vectors refer to the signals from and to the behaviours
respectively (figure 2). Therefore, each unit is an
automaton. Because of this combination of behaviours
and tasks expressed as state machines we call the
architecture hybrid.The described behavioural architecture provides
advantages by its intrinsic features like concurrency of
several behaviours, simplicity of design, abstraction
hierarchy, time response hierarchy, a physically
grounded system [ 11.
VISION IN THE BEHAVIOURAL ARCHITECTURE
The integration of vision in the behavioural
architecture requires operations at all three levels:
defining vision devices at level 1, defining visionbased
behaviours at level 2 and, at level 3, defining the
tasks in term of behaviours, making best use of the
visual behaviours available. Let us consider the three
levels successively.
At level one, defining vision devices consists merely in
selecting a number of vision devices needed or useful
for the application. Vision devices that have been
considered so far in our investigations belong to active
and passive vision, landmark vision, laser range
imaging, sonar and infrared ranging.
At level two and of central concern is the definition of
vision-based behaviours. Each behaviour is defined
and developed as a widely autonomous unit
responsible for a stereotyped action the robot performs
under the control of a stimulus. In vision-based
behaviours the stimulus is a visual pattern. Upon
detection the stimulus initiates a robot action and
maintains it as long as it exists, building up a control
loop with feedback across the environment.
The sorts of behaviours to be implemented are selected
according to sensor capabilities and application
requirements. Typical behaviours considered for
vision-based navigation are: Going tou~nrifsG, oin8 d ong
right, Obstacle avoidance, Obstncle detection, Lnnilrnnrk
following, Wonder nrounii, Hornins, Self-yositionin<ye, tc.
Finally, tasks the robot has to perform are defined at
level three. Each task is packaged in a unit. Typicdl
tasks considered so far are: exploration, navigation,
postman task, tidying up chairs in a room [6][S].
The behavioural approach is based on the existence of
individual behaviours and on the coordination of
these behaviours. It states: autonomy emerges from
the co-operative work of various behaviours [l].
We define a behaviour as an independent stereotyped
action that is maintained by a specific perceived
stimulus. Examples of robot navigation behaviours are
Wonder arortnd or Go along. Each behaviour is activated
by a stimulus. The overall agent behaviour emerges
from the coordination of the various active behaviours.
The behavioural approach is of special interest for
building agents which interact strongly with their
environment by sensing.
Figure 2 illustrates the basics of a the behavioural
architecture which derives from this approach. We
recognise a structure organised in three levels of
abstraction:
Level 1: physical
Level 2: behavioural
Level 3: cognitive
Level 1 includes all devices for acting on the robot and
for sensing the environment.
Level 2 is the heart of the architecture and implements
the behaviours as independent modules.
Level 3 includes units which implement the tasks to be
performed by the agent. The units proceed by
activation/deactivation of the behaviours. Units can be
built according to very different principles [6].
In the frame of this paper, we select the principle of a
task described by a state machine whose transitions
are controlled by a status vector and each state gives
rise to an activation vector. Status and activation
vectors refer to the signals from and to the behaviours
respectively (figure 2). Therefore, each unit is an
automaton. Because of this combination of behaviours
and tasks expressed as state machines we call the
architecture hybrid.The described behavioural architecture provides
advantages by its intrinsic features like concurrency of
several behaviours, simplicity of design, abstraction
hierarchy, time response hierarchy, a physically
grounded system [ 11.
VISION IN THE BEHAVIOURAL ARCHITECTURE
The integration of vision in the behavioural
architecture requires operations at all three levels:
defining vision devices at level 1, defining visionbased
behaviours at level 2 and, at level 3, defining the
tasks in term of behaviours, making best use of the
visual behaviours available. Let us consider the three
levels successively.
At level one, defining vision devices consists merely in
selecting a number of vision devices needed or useful
for the application. Vision devices that have been
considered so far in our investigations belong to active
and passive vision, landmark vision, laser range
imaging, sonar and infrared ranging.
At level two and of central concern is the definition of
vision-based behaviours. Each behaviour is defined
and developed as a widely autonomous unit
responsible for a stereotyped action the robot performs
under the control of a stimulus. In vision-based
behaviours the stimulus is a visual pattern. Upon
detection the stimulus initiates a robot action and
maintains it as long as it exists, building up a control
loop with feedback across the environment.
The sorts of behaviours to be implemented are selected
according to sensor capabilities and application
requirements. Typical behaviours considered for
vision-based navigation are: Going tou~nrifsG, oin8 d ong
right, Obstacle avoidance, Obstncle detection, Lnnilrnnrk
following, Wonder nrounii, Hornins, Self-yositionin<ye, tc.
Finally, tasks the robot has to perform are defined at
level three. Each task is packaged in a unit. Typicdl
tasks considered so far are: exploration, navigation,
postman task, tidying up chairs in a room [6][S].
