Jump to the Theo Overview page.
Jump to the Comparative Capabilities page.
Jump to the Home page.

Capabilities

Overview: Theo is an example of a Plan-Then-Compile architecture which integrates learning, planning, and knowledge representation into a single compilation learning system. All of the knowledge of the system is composed of frames, including various examples of meta-knowedge, such as certainty and expected cost.


 
Planning

Planning in Theo is used to infer new stimulus-response behavior rules which it
can store with its current set of reactive rules. 
Theo's planning capability is invoked when the none of the preconditions 
for this current set apply.  That is, Theo plans only 
when necessary.


There is no particular planner which is implicit in the Theo architecture. 
The frame-based nature of the planner can facilitate a STRIPS planner just as
easily as a less linear planner.  In this manner, different planning 
algorithms can be tested by Theo.


When Theo is forced to plan, a new rules is learned to cover the recently 
planned decision, such that the rule will make the same decision as the
planner.  The learning in Theo can therefore be described as impasse-driven
learning.


 
Meta-Reasoning

The knowledge base in theo includes a great amount of metaknowledge, including:



This metaknowledge is to be utilized by the learning 
process which infers new reactive rules which are relevant to the current
environment.





Reactivity

Theo is equipped with a reactive decision maker, such that it can quickly 
react to changes in a dynamic environment.  It does this with the use of
both eager and lazy sensing.  Eager sensing is used in order to make the 
system reactive; it constantly indexes a set of simulus-response rules, and
when the preconditions for one is met, it acts. Lazily sensed features 
however are only sensed when needed.  This keeps the system focused, because
less sensing equates to less thinking, which frees our limited resources.



 
Taskability

Though the various Theo agents at present do not interact with external
sources in order to acquire new goals, this is not an inherent characteristic
of the Theo architecture.  Adding a new directive which prompts a user for
input whenever some given production is used is alluded to be a simple task
in this architecture.




 
Learning

Planning in Theo is used to infer new stimulus-response behavior rules which it
can store with its current set of reactive rules. 
Theo's planning capability is invoked when the none of the preconditions 
for this current set apply.  That is, Theo plans only 
when necessary.


There is no particular planner which is implicit in the Theo architecture. 
The frame-based nature of the planner can facilitate a STRIPS planner just as
easily as a less linear planner.  In this manner, different planning 
algorithms can be tested by Theo.


When Theo is forced to plan, a new rules is learned to cover the recently 
planned decision, such that the rule will make the same decision as the
planner.  The learning in Theo can therefore be described as impasse-driven
learning.


 
Interruptability

Theo is equipped with a reactive decision maker, such that it can quickly 
react to changes in a dynamic environment.  It does this with the use of
both eager and lazy sensing.  Eager sensing is used in order to make the 
system reactive; it constantly indexes a set of simulus-response rules, and
when the preconditions for one is met, it acts. Lazily sensed features 
however are only sensed when needed.  This keeps the system focused, because
less sensing equates to less thinking, which frees our limited resources.
In addition, the learning capability of Theo is
used to constantly update and revise the behavioral rules.


Various Theo robotic agents have been implemented which illustrate the 
architecture's versatility in a dynamic real-world
situation.


 
Navigation/Manipulation

Theo is equipped with a reactive decision maker, such that it can quickly 
react to changes in a dynamic environment.  It does this with the use of
both eager and lazy sensing.  Eager sensing is used in order to make the 
system reactive; it constantly indexes a set of simulus-response rules, and
when the preconditions for one is met, it acts. Lazily sensed features 
however are only sensed when needed.  This keeps the system focused, because
less sensing equates to less thinking, which frees our limited resources.
In addition, the learning capability of Theo is
used to constantly update and revise the behavioral rules.


Various Theo robotic agents have been implemented which illustrate the 
architecture's versatility in a dynamic real-world
situation.


 
Coherent Behavior

Theo is equipped with a reactive decision maker, such that it can quickly 
react to changes in a dynamic environment.  It does this with the use of
both eager and lazy sensing.  Eager sensing is used in order to make the 
system reactive; it constantly indexes a set of simulus-response rules, and
when the preconditions for one is met, it acts. Lazily sensed features 
however are only sensed when needed.  This keeps the system focused, because
less sensing equates to less thinking, which frees our limited resources.
In addition, the learning capability of Theo is
used to constantly update and revise the behavioral rules.


Various Theo robotic agents have been implemented which illustrate the 
architecture's versatility in a dynamic real-world
situation.



Perception

Theo is equipped with a reactive decision maker, such that it can quickly 
react to changes in a dynamic environment.  It does this with the use of
both eager and lazy sensing.  Eager sensing is used in order to make the 
system reactive; it constantly indexes a set of simulus-response rules, and
when the preconditions for one is met, it acts. Lazily sensed features 
however are only sensed when needed.  This keeps the system focused, because
less sensing equates to less thinking, which frees our limited resources.
In addition, the learning capability of Theo is
used to constantly update and revise the behavioral rules.


Various Theo robotic agents have been implemented which illustrate the 
architecture's versatility in a dynamic real-world
situation.



Other