Question

1[30%]. This problem is similar to question 2.10 in the text book, except the environment configuration is slightly different. Consider the following vacuum-cleaning world which has 3 squares: We assume the following: The vacuum-cleaning agent knows the configuration of the 3 squares, but does not know which square it is in initially. The initial dirt distribution is not known either. A clean square remains clean and a dirty square remains dirty unless the agent cleans it up. The agent has two sensors: Loc(x) and Status(x). Loc(x) returns the identity ("A" or "B" or "C") of the square x that the agent is located, and Status(x) returns either "Dirty" or "Clean" for the square x. Assume that the sensors are 100% reliable. The actions available for the agent are: Left, Right, Suck, NoOp. Each action takes place in one time "step". The actions are reliable - performing "Suck" in dirty square will change it to become clean, and moving left when in square C will get the agent to square B. The agent will be penalized by -1 point for each movement Left, Right, Suck. The agent gets +10 points reward for each dirty square cleaned. For a time horizon (say T = 20) steps, the agent is penalized by -2 points per dirty square per time step. (If a square is dirty at the "start" of time step t and the agent performs "Suck" at this time step, the square becomes "clean" at the "end" of step 1. Thus for time step for this particular square, no penality should be applied to the agent). The agent's performance is measured by the total number of points (positive or negative) over T = 20 time steps. pre renlex argent (without states, To M e . the went and define the "condition-action" rules. Can such an awantu. p erTICASLICIORCY. di (b) What about a reflex agent with states (i.e., an agent which has an internal state representation of the world environment)? If you like, you can also allow some randomized actions by the agent (in some state). Will this agent perform better than the simple reflex agent in part (a)? Briefly explain your answer. Design such an agent (pseudo-code, rules, or look-up table, etc.). som CURSOS etc.).

Answer 1

A simple reflex agent does an action if a condition is given to it. It does not have a wide spectra of knowledge. So we need to give the correct commands and it will perform accordingly. From the above we know that the vacuum cleaner is an example of this kind of agent. We have to feed it environment knowledge as conditions. So let us see a simple pseudo code with which you can develop the operations of the vacuum cleaner.

Pseudo code for vacuum operations:

1.Assume the position of the agent in an array form. Since there are three spaces given we set an array n holding 3 positions.

2.Declare variables for status as d for dirty and c for clean as boolean values.

3.Declare a variable to store the points value.

4.If particles are present the agent can sensor them and the space is dirty.

5.If d variable becomes true then perform the suck operation.

6.Set a count variable to see how many moves the agent makes. Set a random count value and say that the agent can make

these many counts in the specific time(20).

7.Declare the time taken should be 20 and if it takes more time than that then agent fails in completing the task.

8.Use a for loop to iterate the positions. i.e n=0 and so. If n=2 then n-1 is done to go to the previous position. If agent is in C space then (n-1) is done so that it goes to B position

9.If c variable becomes true then that means the room is clean and NoOp is done and the agent stops the task and exits the loop.

10.Finally calculate the performance of the agent by multiplying the point achieved and time which is already given i.e. 20.