Logic-Based Planning

We can represent the Wumpus rules in a Logical System (e.g. Propositional Logic, First-Order Predicate Logic, ALC) and use Inference to deduce new knowledge from percepts and actions. This representation will result in contradictions when the agent moves through the world and its percepts are changing.

The obvious idea is to make the representations of percepts time-dependent. Fluent, Atemporal.

• Model-based Reflex Agent
• Knowledge Base Agent

We want an Agent that can deal with time. For example we would like to be able to express something like this: $\forall t,x,y.\mathrm{Ag}@(t,x,y)\Rightarrow \mathrm{draft}(t)\iff \mathrm{breeze}(x,y).$ But how do we get a Fluent like: the agent is at $x,y$ at time $t$ ? Has to do with Transition Model where we get the state of the Environment after an Action (so for example new cell position).

We can use fluents to describe actions in so called Effect Axioms. For example:

Frame Problem

Hybrid Agent → special code for Wumpus world example.

However, we want: General Planning Agent.