Information Technology Reference
In-Depth Information
510 Advanced Artificial Intelligence
8. get-new-external-events()
9. drop-successful-attitudes(B,G,I)
10. drop-impossible-attitudes(B,G,I)
11. until quit
14.3.3
Reactive agent
The problems in traditional artificial intelligence are reflected in deliberative
agent without any change. The main criticism focuses on the rock-bound
architecture. The agents work in dynamic environment. So, they need to make a
decision according to the current conditions. However, their intention and plan
were developed on the symbol model of past specific time. There is little change
about that. Rule-based rock-bound extends the disadvantage since the conversion
between planer, scheduler and executor is time cost. The implementation of
condition of scheduler changes more or less. The symbol algorithms of
deliberative agents are often ideal and decidable which cause the high
complication. In dynamic environment, it is more important to meet the
requirement than the plan optimization. On other way around, deliberative agents
are good at mathematics proof of plans.
Different with the deliberative agents, Reactive agents include no world
model. They have not complex symbol reasoning (Wooldridge, 1995). Figure
14.8 give the architecture of reactive agents. The condition-action rules connect
the cognition and actions. The rectangle in the figure represents the current inner
states of decision process. The ellipse shows the background knowledge which is
used in the process.
Algorithm 14.4
Reactive-agent algorithm.
Reactive-agent (percept) returns action
static: state, describe the current world state
rules, a group of condition-actions rule
1. state ← Interpret-Input(percept)
2. rule ← Rule-Match(state, rules)
3. action ← Rule-Action[rule]
4. return action
In the program, Interpret-Input function generates the abstract description of
current states from the percept. Rule-Match function returns a rule which is
matched with the description of current states.
