Recent Changes

• A "Discussion" was added to Problem 2.8 below (01/29/08, 9:30am)

Problem 2.3

Problem 2.4

Here is a state machine modeling the nest preparation of the lowly dung beetle, as described on page 37 of R&N:

Problem 2.7

• the indicated circuit from the book, which latches (and continues to ring) if the vehicle has seen a light in at least one previous time instance
• one of your own design that latches/rings if the vehicle has seen a light in at least two different time intances
• one of your own design that continues to ring if the vehicle has seen a light in exactly two different time instances. Hint: start with the one you just designed, but inactivate the bell with inhibition from some additional circuitry that latches whether a light has been seen in at least three time instances

Problem 2.8

Note: The book's code repository might not include your language of choice. In this case, you must write your own. However, you can certainly look at (or even translate) the implementations given, considering them very detailed pseudo-code.

The book website is http://aima.cs.berkeley.edu/ About halfway down in the left column is the "Online Code Repository"

Discussion

• Overview. You need to write a simulator for a stimulus-response agent (of your design) acting in the 1x2 world. You need to compute its performance measure (of your design) for each of the 8 possible worlds.
• The simulation. Your simulator need only work for a stimulus-response agent acting in the 1x2 world. Your simulator should be able to take an arbitrary environment (of the eight possible) as the start location. In that case, you would have to run it 8 times to gather all 8 performance measures for your agent. Alternatively, you could easily loop over all environments, simulating your agent in each one to obtain its performance measure in that environment.
• The S/R agent. You need to make an STIMULUS-RESPONSE agent, not an arbitrary. For example, my S-R agent (yours can vary) is the following

  A, Clean --> Left
  A, Dirty --> Left
  B, Clean --> Left
  B, Dirty --> Left
  

  If this agent started in room B, with both squares initially dirty, it would move to room A after the first step and then (trying to move Left from there on each step, but running into the wall each time) stay there for as many steps as the simulation lasted.

• The performance measure. This is not stated and is up to you. An example of one that makes sense to me for judging/comparing different S/R agents in this environment is as follows: simulate the agent for 10 steps, penalizing it 1 point for each square that is dirty in each time step. So, for the initial state and agent mentioned in the previous point, the performance would be -2 per time step (both rooms are dirty for each time step).
• Write-up. Your writeup definitely needs to detail the S/R agent you used as well as the performance measure you used. You also must report the the performace measure on each of the environemts. Finally, it would be appropriate to BRIEFLY explain a little about how your simulator code works (data structures, looping, etc.)