onboard controllers is message-based. Commands and responses are in
The most traditional robot control architecture is structured in layers of
functional modules as depicted in Figure 10.6. There is a straightforward
transformation of sensor data (perception) in environment representation
(modelling), decision-making (planning), task decomposition (execution),
and elementary commands (motor control).
Several robot control architectures have been derived as refinements or
extensions of this control-feedback loop and have been documented in the
Reactive architectures minimize the role of modelling and planning ac-
tivities in order to cope with real time requirements of the robot controller.
The controller reacts to external stimuli, such as the presence of an obstacle
along its path, by exhibiting elementary behaviours, such as stopping the
motors or turning the robot in the opposite direction.
Deliberative architectures rely on a detailed description of the surround-
ing environment in order to enable the robot to plan its activities and carry
out useful tasks. Tasks are described in terms of goals that the robot has to
achieve, such as transporting an object from one place to another (e.g. two
rooms in a building).
Hybrid architectures split the control activity into a number of inde-
pendent behaviours that enforce the robot's reactivity to external stimuli
and goal directedness. Typically they are decentralized architectures, i.e.
they are made up of a collection of autonomous and collaborating control
Subsumption architectures are special types of hybrid architectures,
where reactive behaviours (such as landmark detection or obstacle avoid-
ance) and deliberative behaviours (such as map building and navigation)
use sensor data independently and asynchronously to produce control com-
mands. Reactive and deliberative behaviours have the ability to inhibit or
suppress the control commands of other behaviours.
Figure 10.6 Traditional robot control feedback loop