92
Biologically Inspired Robotics
pattern is optimized to achieve the best performance. In the behavior layer,
behaviors are optimized according to the encoding method of behaviors. If
behaviors are encoded by fuzzy logic controllers, fuzzy rulers and fuzzy
function parameters will be optimized. In the cognitive layer, optimization
also could be designed for the parameters of reasoning and planning tasks
(Liu, Hu, and Gu 2006).
5.4.1 Cognitive Layer
The cognitive layer extracts the robotic fish status from the sensor states and
then performs task-oriented reasoning and planning. It makes decisions
on how to coordinate fish behaviors to achieve a given task. Rather than
providing a detailed objective for the behavior layer, for example, a particular trajectory for the robotic fish to follow, the cognitive layer adjusts the
contribution weights of individual behaviors for the emergent behavior. In
addition, the cognitive layer will adjust the parameters for some individual
behaviors, such as the depth level for keep-level behaviors.
In a known environment with available models, planning can be done
offline, and solutions can be found and evaluated prior to execution.
However, in dynamically changing environments, models and policies
need to be adaptively revised online at this layer. Trial-and-error processes
are normally carried out. As can be seen from Figure 5.6, the raw sensor
data are converted into sensor states so that individual behaviors take sensor states as stimulus and calculate responses concurrently. The cognitive
layer extracts the abstracted states from the sensor states. The parameters for
behavior coordination are transferred into the behavior layer after reasoning
and planning.
5.4.2 Behavior Layer
The behavior layer is located in the middle of the proposed layered control
architecture for the reactive control of our robotic fish. It firstly converts the
sensor raw data into sensory states by fuzzy membership functions and
then inputs these states into individual behaviors. Fuzzy logic controllers
are designed for individual behaviors to process these states. The responses
of all behaviors are coordinated together by the behavior coordination
function, which is defined by the parameters from the output of the cognitive layer. The subsumption architecture is adopted for this layer because
of its simplicity. Here the stimulus is the sensor states and the response is
the action. Stimulus–response (SR) diagrams (Arkin 1998) are used for the
design of specific behavioral configurations. Eight behaviors are designed
for the generic purpose in this layer, namely:
Biologically Inspired Robotics
pattern is optimized to achieve the best performance. In the behavior layer,
behaviors are optimized according to the encoding method of behaviors. If
behaviors are encoded by fuzzy logic controllers, fuzzy rulers and fuzzy
function parameters will be optimized. In the cognitive layer, optimization
also could be designed for the parameters of reasoning and planning tasks
(Liu, Hu, and Gu 2006).
5.4.1 Cognitive Layer
The cognitive layer extracts the robotic fish status from the sensor states and
then performs task-oriented reasoning and planning. It makes decisions
on how to coordinate fish behaviors to achieve a given task. Rather than
providing a detailed objective for the behavior layer, for example, a particular trajectory for the robotic fish to follow, the cognitive layer adjusts the
contribution weights of individual behaviors for the emergent behavior. In
addition, the cognitive layer will adjust the parameters for some individual
behaviors, such as the depth level for keep-level behaviors.
In a known environment with available models, planning can be done
offline, and solutions can be found and evaluated prior to execution.
However, in dynamically changing environments, models and policies
need to be adaptively revised online at this layer. Trial-and-error processes
are normally carried out. As can be seen from Figure 5.6, the raw sensor
data are converted into sensor states so that individual behaviors take sensor states as stimulus and calculate responses concurrently. The cognitive
layer extracts the abstracted states from the sensor states. The parameters for
behavior coordination are transferred into the behavior layer after reasoning
and planning.
5.4.2 Behavior Layer
The behavior layer is located in the middle of the proposed layered control
architecture for the reactive control of our robotic fish. It firstly converts the
sensor raw data into sensory states by fuzzy membership functions and
then inputs these states into individual behaviors. Fuzzy logic controllers
are designed for individual behaviors to process these states. The responses
of all behaviors are coordinated together by the behavior coordination
function, which is defined by the parameters from the output of the cognitive layer. The subsumption architecture is adopted for this layer because
of its simplicity. Here the stimulus is the sensor states and the response is
the action. Stimulus–response (SR) diagrams (Arkin 1998) are used for the
design of specific behavioral configurations. Eight behaviors are designed
for the generic purpose in this layer, namely: