ARCHITECTURE FOR INTELLIGENT

AGENTS
 Logic-Based Architectures
• Intelligent behavior can be generated in a system by giving that system a
symbolic representation of its environment and its desired behavior.
• These symbolic representations are logical formulae, and the syntactic
manipulation corresponds to logical deduction, or theorem proving.
• Such agents are assumed to maintain an internal database of formulae of
classical first-order predicate logic, which represents the information they
have in symbolic form.
• For example, an agent’s belief database might contain formulae such as the
following:
✔ Open(valve)
✔ Temperature(reactor,321)
✔ Pressure(tank,28)
• The database is the information that the agent has about its environment.
• An agent’s decision-making process is modeled through a set of deduction
rules.
• The action takes as input the beliefs of the agent (Δ) and deduction rules (ρ)
and returns as output either an action or else null when nothing is found
Figure 1: Vacuum Cleaning World Figure 2: Simple Agent Architecture
• To illustrate these ideas, let us consider a small example based on the vacuum
cleaning world example. The idea is that we have a small robotic agent that
will clean up a house.
• The robot is equipped with a sensor that will tell it whether it is over any dirt,
and a vacuum cleaner that can be used to suck up dirt.
• In addition, the robot always has a definite orientation (North, East, West and
South) and turn right 90ᴼ.
• The agent moves around a room, which is divided grid-like into a number of
equally sized squares.
• We will assume that our agent does nothing but clean – it never leaves the
room.
• Our agent can receive a percept dirt, or null.
• It can perform any one of three possible actions: forward, suck, or turn.
• The robot will always move from (0,0) to (0,1) to (0,2) and then to (1,2), to
(1,1), and so on.
• The goal is to traverse the room, continually searching for and removing dirt.
• First, make use of three simple domain predicates:
✔ In(x, y) agent is at (x,y)
✔ Dirt(x, y) there is dirt at (x,y)
✔ Facing(d) the agent is facing direction d
•The rules that govern our agent’s behavior are:
•The problems with this vacuum cleaning world are
Calculative rationality is clearly not acceptable in environments that change
faster than the agent can make decisions.
Representing and reasoning temporal information. Temporal information is
how a situation changes over time. Representing it turns out to be
extraordinarily difficult.
The problems associated with representing and reasoning about complex,
dynamic, possibly physical environments are also essentially unsolved
 Reactive Architectures (or) Subsumption
Architecture
• The subsumption architecture is arguably the best-known reactive agent
architecture.
• There are two defining characteristics of the subsumption architecture.
❖ The first is a set of task accomplishing behaviors, each behavior may be
thought of as an individual action selection process, which continually takes
perceptual input and maps it to an action to perform.
❖ The second defining characteristic of the subsumption architecture is that
many behaviors can “fire” simultaneously.
• A subsumption hierarchy has behaviors arranged into layers.
• Lower layers in the hierarchy are able to inhibit higher layers:
• the lower a layer is, the higher is its priority.
For example, in a mobile robot the behavior avoid obstacles. It makes sense to give
obstacle avoidance a high priority.
• The objective is to explore a distant planet, more concretely, to collect samples
of a particular type of precious rock. The location of the rock samples is not
known in advance, but they are typically clustered in certain spots. A number of
autonomous vehicles are available that can drive around the planet collecting
samples and later reenter the mothership spacecraft to go back to earth. There
is no detailed map of the planet available, although it is known that the terrain
is full of obstacles – hills, valleys, etc. – which prevent the vehicles from
exchanging any communication.
The solution makes use of two mechanisms introduced by Steels.
❖ The first is a gradient field, the range of radio signal generated by
mothership to find its location.
❖ The second is communicate mechanism. agents will carry radioactive
crumbs, which can be dropped, picked up, and detected by passing
robots.
• The lowest-level behavior is obstacle avoidance, which can be represented in
the rule:
if detect an obstacle then change direction ------- 1
• Other behaviors ensures any samples carried by agents are dropped back at
mothership.
if carrying samples and at the base then drop samples.------ 2
if carrying samples and not at the base then travel up gradient.---- 3
if detect a sample then pick sample up. --------- 4
if true then move randomly.----------- 5
• These behaviors are arranged into the following hierarchy:
1<2<3<4<5
• However, rule 3, determining action on carrying sample and not at base is
modified as follows.
if carrying samples and not at the base then drop 2 crumbs and travel up
gradient. -------- 6
• However, an additional behavior is required for dealing with crumbs.
if sense crumbs then pick up 1 crumb and travel down gradient.---- 7
• These behaviors are then arranged into the following subsumption hierarchy:
1<3<6<4<7<5
• Advantages and Disadvantages of Reactive Architecture:
Advantages to reactive approaches with Brooks’s subsumption architecture
❖ simplicity
❖ economy,
❖ computational tractability,
❖ robustness against failure.
Disadvantages
❖ Agents need sufficient information available in their local environment.
❖ It is difficult to see how decision making could take into account non-local
information.
❖ Overall behavior emerges from the interaction of the component behaviors.
But relationship between individual behaviors, environment, and overall
behavior is not understandable.
❖ It is much harder to build agents that contain many layers. The dynamics of
the interactions between the different behaviors become too complex to
understand.
 Belief-Desire-Intention Architectures
•the process of deciding, moment by moment, which action to
perform in the furtherance of our goals.
•Practical reasoning involves two important processes:
Deliberation - deciding what goals we want to achieve, and
Means-ends reasoning - how we are going to achieve these goals.
Intention
Make a reasonable attempt to achieve the intention.
if a course of action fails to achieve the intention, then you would
expect to try again – you would not expect to simply give up. This
intention will constrain future practical reasoning.
•Intentions play a number of important roles in practical reasoning:
Intentions drive means-ends reasoning.
Intentions constrain future deliberation.
Intentions persist.
Intentions influence beliefs upon which future practical reasoning is based.
The rate of world change is γ.
If γ is low (i.e., the environment does not change quickly) then bold agents
do well compared to cautious ones, because cautious ones waste time
reconsidering their commitments while bold agents are busy working
towards – and achieving – their goals.
If γ is high (i.e., the environment changes frequently) then cautious agents
tend to outperform bold agents, because they are able to recognize when
intentions are doomed, and also to take advantage of serendipitous
situations and new opportunities.
•Practical Reasoning
A set of current beliefs, representing information agent has about
its current environment;
•A belief revision function (brf), which takes a perceptual input and
the agent’s current beliefs, and on the basis of these, determines a
new set of beliefs;
•An option generation function(options), which determines the
options available to the agent (its desires), on the basis of current
beliefs and its current intentions;
•A set of current options, representing possible courses of actions
available to the agent;
•A filter function (filter), which represents the agent’s deliberation
process, and which determines the agent’s intentions on the basis of
its current beliefs, desires, and intentions;

•A set of current intentions, representing the agent’s current focus –

those states of affairs that it has committed to trying to bring about;

•An action selection function (execute), which determines an action

to perform on the basis of current intentions.
•The state of BDI agent at any given moment is a triple (B,D,I), where
B ⊆ Bel, D ⊆ Des, and I ⊆ Int.
•If we denote the set of possible percepts that the agent can receive
by P, then
• Thus filter should satisfy the following constraint:
Pseudo-code of function action
 Layered Architectures
•Layered Architectures involves creating separate subsystems, as a
hierarchy of interacting layers, to deal with reactive and proactive
behaviors.

•Two examples of such architectures: INTERRAP and

TOURINGMACHINES. There are two types of control flow within
layered architectures.
•Horizontal layering
In horizontally layered architectures, the software layers are each
directly connected to the sensory input and action output.
In effect, each layer itself acts like an agent, producing suggestions
on action to perform.
Advantage – Simplicity.
Drawback - overall behavior of the agent will not be coherent.
•Vertical layering
In vertically layered architectures, sensory input and action output
are each dealt with by at most one layer each.
A mediator function makes decisions about which layer has
―control of the agent at any given time.
Drawback - designer must potentially consider all possible
interactions between layers.
•Vertically layered architecture is subdivided into one-pass
architectures and two-pass architectures.

•The complexity of interactions between layers is reduced in vertical

architecture.

•Since there are n−1 interfaces between n layers, then if each layer is
capable of suggesting m actions, there are at most m pow 2 (n−1)
interactions to be considered between layers.
•Touring Machines
The Touring Machines architecture consists of perception and action
subsystems, which interface directly with the agent’s environment, and
control layers embedded in a control framework, which mediates between
the layers.
TOURING MACHINES consists of three activity producing layers.
✔ Reactive layer: immediate response.
✔ Planning layer: ―day-to-day running under normal circumstances.
✔ Modelling layer: predicts conflicts and generate goals to be achieved in
order to solve these conflicts.
✔ Control subsystem: decided which of the layers should take control over
the agent.
•INTERRAP
INTERRAP defines an agent architecture that supports situated behavior
where the agents are able to recognize unexpected events and react timely
and appropriately to them.
It show goal-directed behavior in a way that the agent decides which goals
to pursue and how.
The agents can act under real time constraints and act efficiently with
limited resources.
The agent can interact with other agents to achieve common goals.
INTERRAP - a horizontal layered two-pass agent architecture
INTERRAP - a vertically layered two-pass agent architecture.
 AGENT COMMUNICATION
•Multi-agent System (MAS) are distributed systems. They are
collection of collaborative agents.
•Engineering a multi-agent system means rigorously specifying the
communications among the agents by way of interaction protocols.
•The agents are autonomous and heterogeneous entities and
applications are auctions, banking, shipping, and so on
•Autonomy and Its Implications
Each agent is an autonomous entity in the sense that agent itself
is a domain of control, other agents have no direct control over
its actions.
Protocols are modular, potentially reusable specifications of
interactions, called messages, between two or more
components.
A protocol is designed with a certain application in mind. An
enactment refers to an execution of the protocol by the
components.
As long as components are individually conformant, that is, follow
their respective roles in the protocol, they will be able to work
together no matter how they are implemented.
Interoperation means that the components are loosely coupled with
each other; that is, we can potentially replace a component by
another conformant one and the modified system would continue to
function.
❖The irrelevance of intelligence
✔ the ability of an agent to perform high-level reasoning or as the
degree to which an agent can operate without the supervision of its
principal. Two agents may act on two different worlds.
❖Logical versus physical distribution
✔ Because of their autonomy, agents are the logical units of
distribution, where two or more agents interacts.
✔ The choice of whether an application is implemented as a single
process or multiple ones is often driven by physical considerations
such as geographical distribution, throughput, redundancy, number of
available processors and cores, and so on.
• Heterogeneity refers to the diversity of agent implementations. A component
can be implemented based on its dependence on other components, without
concern for how the others are implemented.
• Accommodating heterogeneity entails specifying the semantics of the
interaction.
• Making an offer in a Multi-agent system is social commitment.
Fig 1: Updating an offer
•Criteria for Evaluation
❖ Software engineering
✔ Protocol specifications should be modifiable, easily understandable, and
composable.
❖ Flexibility
✔ Flexibility is especially important in dynamic settings where agents may
come and go, and exceptions and opportunities may arise.
❖ Compliance checking
✔ Compliance checking means determining if the agent is following the
protocol.
• Communication in Multi Agent System
The main insight of Speech act theory, better known as communicative act
theory, is that communication is a form of action.
Performative verbs such as, request, promise, etc., makes the sentences
stylish.
Traditional Software Engineering Approaches
Sequence Diagrams
The most natural way to specify a protocol is through a message sequence
chart (MSC), formalized as part of UML as sequence diagrams.
each edge connecting two lifelines indicates a message from a sender to a
receiver.
Time flows down ward with the orderings of the messages.
MSCs support primitives for grouping messages into blocks.
• FIPA (Foundation of Intelligent Physical Agents) is a standards body, now part
of the IEEE, has formulated agent communication standards.

FIPA request interaction protocol,

expressed as a UML sequence
diagram