Ambient Intelligence Through Agile Agents - Ambient Intelligence for Scientific Discovery - page 287

computationally challenged devices which will typify ambient scenarios. This has
been achieved by ensuring that the Run-Time Environment, which includes the
AF-APL Interpreter, is compliant with version 1.1.8 of the Java SDK (a.k.a.
Personal Java for Mobile Devices). To check compatibility with future versions
of Java, J2ME-compliant versions of the Run-Time Environment have also been
developed. However, due to incompatibilities between Personal Java and J2ME,
and as a result of our wish to ensure that AF can be deployed on the most
prevalent operating system and JVM configuration for PDAs (e.g.. MS PocketPC
and Jeode), AF is currently not J2ME-compliant.

2.1 AF-APL

AF-APL is a declarative Agent-Oriented Programming (AOP) language that
supports the programming of agent behaviors. The basic premise behind AF-
APL is the view that complex agent behaviors can be more naturally modeled by
viewing agents to be mental entities that maintain an internal mental state which
is comprised of mental attitudes, in this case: beliefs and commitments. Beliefs
describe, using a first-order logic representation language, the current state of the
agent and its environment, and commitments describe the current (and future)
activities that the agent has decided to perform. Finally, decisions are modeled
through a set of commitment rules that map situations (a conjunction of positive
and negative beliefs) onto commitments. These rules are checked repeatedly
within a sense-deliberate-act cycle.
Beliefs represent the current state of both the agent and its environment.
In AF-APL, this state is realized as a set of facts that describe atomic infor-
mation about the environment, and which are encoded as first-order structures
wrapped within a belief operator ( BELIEF ). For example, in a mobile computing
application an agent may be asked to monitor the users current position using a
Global Positioning System (GPS) device. The agent may generate a belief about
this position that takes the form: BELIEF(userPosition(Lat, Long)) where
Lat and Long are replaced by values for the user's latitude and longitude re-
spectively. In AF, the actual values for the latitude and longitude are retrieved
directly from the GPS device by a perceptor unit, which converts the raw sensor
data into corresponding beliefs. The triggering of this perceptor unit is part of
a perception process, which is central to our strategy for updating the beliefs
of agents and is realized by triggering a pre-selected set of perceptor units at
regular intervals. The specific set of perceptor units to be used by an agent is
specified as part of the agent program through the PERCEPTOR keyword.
Commitments represent the courses of action that the agent has chosen to
perform. That is, they represent the results of some reasoning process in which
the agent makes a decision about how best to act. From this perspective, com-
mitment implicitly represents the intentions of the agent. This contrasts with
more traditional BDI approaches [32, 37] in which intention is represented ex-
plicitly and commitment is an implicit feature of the agents' underlying reasoning
process. This alternative treatment of commitment is motivated by our goal of
explicitly representing the level of commitment the agent has to a chosen course