I have defined a Basic-SNePS-wumpus-agent that operates in the wumpus world. It is based on the aimless-wumpus-agent, defined in the file /projects/shapiro/AIclass/Aimacode/agents/agents/wumpus.lisp, and on the material in Chapters 6 and 7 of our text. The Basic-SNePS-wumpus-agent works, but is still pretty stupid. For Project 3, you are to modify it to be smarter.

Here are some suggested ways to make the agent smarter. For full credit, make all these modifications, or an equal number of equivalently interesting modifications.

• On page 210, the text discusses various kinds of actions. The current agent makes the great actions of grabbing the gold when appropriate, and climbing out of the cave when it has the gold and is at the home square. It also makes the good action of going to a safe, unvisited square, and the medium action of moving to an already visited square, which must be safe (why?). However, it never knowingly makes the risky action, and this sometimes causes it to just move back and forth among a very few squares, even though it could get the gold by being less timid. (Does the default random move help here?)

• The current agent shoots its arrow randomly. It should be possible for it, under certain circumstances, to shoot directly at the wumpus, and know it will successfully kill it.

• Once this agent gets the gold, it just continues exploring the cave instead of heading directly for the home square.

The paper, as usual, should discuss your modifications, illustrate the performance of your agent in action, and satisfy all the course requirements for a project paper.

You are to submit your file modified file by 1:30 PM, Tuesday, April 4, 2000, and then turn in your paper at the start of lecture on that same day, Tuesday, April 4, 2000.