E N V I R O N M E N
T S
When we talk about an environment we are describing the
particular situation in which an agent is present.
Agents do not normally have complete control over their
environments but more likely have partial control. This
means that they can influence the environment. In turn,
changes to the environment will directly affect the agents
in the environment.
Environments are generally said to be non-deterministic.
That is to say actions performed in certain environments
can fail. Also, an agent performing the same task in two
identical environments may result in completely different
outcomes.
An agent in an environment will have a pre-programmed set
of abilities that it can apply to different situations that
it may come across in its environment. These abilities are
called effectoric
capabilities. From
the diagram on agents you
can see that the agent has a sensor that is obviously
affected by the environment. The agent can use
information from this sensor along with information
already known in order to decide which action to
perform. Obviously not all actions can be performed in
every situation. For example, an agent may have the
ability to ‘open door’ but can only do so if
the door is unlocked. The fact the door must be unlocked
before the action can be performed is called a
precondition.
The biggest problem associated with agents in an
environment is deciding which action to perform at a given
instance to maximise its chance of completing its goal or
at least working towards it. The complexity of the decision
making process is affected by the properties of an
environment.
Russell
and Norvig suggest the following environment properties:-
•
Accessible/inaccessible:
This refers to the possibility for an agent to obtain
complete and accurate information about the
environment’s state.
Examples:
Inaccessible environment: physical
world: information about any event on earth
Accessible
environment: empty
room which state is defined by its temperature and agents
can measure it.
•
Determinism/non-determinism:
In a
deterministic environment any action has a single
guaranteed effect, and no failure or uncertainty. On the
contrary is a non-deterministic environment. In this
environment, the same task performed twice may produce
different results or may even fail completely.
Examples:
Non-deterministic environment:
physical world: Robot on Mars
Deterministic
environment: Tic
Tac Toe game
•
Episodic/non-episodic:
In an
episodic environment, each agent’s performance is the
result of a series of independent tasks performed. There is
no link between the agent’s performance and other
different scenarios. In other words, the agent decides
which action is best to take, it will only consider the
task at hand and doesn’t have to consider the effect
it may have on future tasks.
Examples:
Episodic environment: mail
sorting system
Non-episodic
environment: chess game
•
Static/Dynamic:
An
environment is static if only the actions of an agent
modify it. It is dynamic on the other hand if other
processes are operating on it.
Examples:
Dynamic environment: physical
world
Static
environment: empty office with no moving
objects
•
Discrete/Continuous:
An
environment is said to be discrete if there are a finite
number of actions that can be performed within it.
Examples:
Discrete environment: A game
of chess or checkers where there are a set number of moves.
Continuous
environment: Taxi
driving. There could be a route from to anywhere to
anywhere else.
The more complex an environment is, the harder it is to
decide which action to perform. The most complex
environment is one that is inaccessible, non-deterministic,
non-episodic, dynamic and continuous.