1. Problem Domain
You will be considering the Wumpus world introduced in Russell and Norvig (2009) Chapter 7. For this programming assignment you'll use Prolog's ability to do inference in order to write an agent that will make safe moves within its world.
2. Project Infrastructure
1) You will need to download and install your own version of SWI-Prolog. It is freely available for major operating systems. Major Linux distributions have copies available within their package system.
2) Download the following three files:
I attached these files.
3) The first two files comprise a simulator for the Wumpus world. The third is an empty agent skeleton program. Your assignment is to flesh out the details in my agent.pl
3. Running the code
Start SWI-Prolog:
pl
Load the world simulator:
?- [wumpus_world].
Load the agent:
?- [my_agent].
Run the agent:
?- evaluate_agent(1, Score, Time).
Correct to: "wumpus:evaluate_agent(1, Score, Time)"? y
If you read the output, you'll see the default agent just moves forward. It does this in the world shown in Russell and Norvig's Figure 7.2, which causes the agent's dead rather soon. (You use Ctrl+D to exit the programming environment.)
You will need to edit my agent.pl produce a more rational movement. To do this you will need to define an intelligent run_agent (Percept, Action). The percepts are provided as a list of five elements [Stench, Breeze, Glitter, Bump, Scream], where you receive a "yes" or a "no" in the respective position.
There is also an init_agent function which you can use for initialization.
The simulator knows the total state of the world, while your agent will know only what it senses through its perceptions. Initially the agent knows only the perceptions in position (1,1) and in order to acquire other information must be moved in other cells of the grid. Your agent must maintain its own state, containing all the information that gradually becomes available. The state will have to be update with the new perceptions every time an action is executed, and this information must be represented so as to permit reasoning (via interference) as to what operations are safe.