A. Since the robot cannot see the dirt on the adjoining cells, we can do nothing but traverse the entire grid either row wise or column wise. Wat we can do is, using the starting position of the robot, we can decide whether it should go up or down, and also whether is should move towards right or left wen the columns are over...

For example, in the given grid

INITIAL: NE(3,5)
MOVES: 30

	S 0.2
	D 0.2
	S 0.4
	D 0.4
	S 0.7
	D 0.7
	S 0.8
	D 0.8
	S 0.9
	D 0.9
	L 0.9
	S 1.1
	U 1.1
	S 1.6
	U 1.6
	S 2.1
	U 2.1
	S 2.9
	U 2.9
	S 3.7
	U 3.7
	S 4.0
	U 4.0
	S 4.5
	U 4.5
	S 4.6
	L 4.6
	S 5.4
	D 5.4
	S 5.9


INITIAL: SW(7,1)
MOVES: 30

	U 0.0
	U 0.0
	U 0.0
	S 0.3
	U 0.3
	S 0.6
	U 0.6
	S 0.7
	U 0.7
	R 0.7
	S 1.2
	D 1.2
	D 1.2
	S 1.7
	D 1.7
	S 1.8
	D 1.8
	D 1.8
	D 1.8
	D 1.8
	R 1.8
	U 1.8
	U 1.8
	U 1.8
	S 2.0
	U 2.0
	S 2.7
	U 2.7
	S 3.1
	U 3.1


B. Since we are using a greedy mechanism, the score this way will be better than the last robot.

INITIAL: NE(3,5)
MOVES: 30

	S 0.2
	U 0.2
	S 0.7
	L 0.7
	S 1.2
	L 1.2
	S 1.7
	U 1.7
	S 2.5
	L 2.5
	S 3.0
	L 3.0
	S 3.0
	D 3.0
	S 3.1
	D 3.1
	S 3.4
	R 3.4
	S 3.9
	R 3.9
	S 4.3
	D 4.3
	S 5.0
	R 5.0
	S 5.8
	D 5.8
	S 6.6
	D 6.6
	S 7.1
	D 7.1
	S 7.6


INITIAL: SW(7,1)
MOVES: 30


C.

D.
a. A simple rexlex agent will be perfectly rational in an environment where all the tiles on the floor are equally dirty (all the values in the grid are same).

b. No, the greedy robot in part b will not be deterministic because it takes random action when the choices are equal, therfore running the robot twice on the same grid may give different results.

c. An environment in which the randomised agent performs poorly can be designed. An example of such an environment is given below.
	
	0	4.0	5.0	4.0	3.0
	0	6.0	0	0	0	
	0	2.0	0	0	0
	2.0	R	2.0	0	0
	0	2.0	0	0	0

Consider the case of such an environment, the robot is at position 'R', it has equal probability of going up, down, right, left (because it goes randomly when it has equal choices) but it is clearly visible that the best result can only be obtained when the robot goes up, so the probability of the best result gets highly reduced, so the robot is performing poorly.

d. If the sensing is noisy, the performance of the rational agent will certainly drop a lot. There will be a good chance that the robot makes a wrong choice at any point thus not performing to its potential. What we can do is, while considerring the values in the grid, we can add the effect of 20% disturbance in the readings, but this ill also not help a lot in improving the performance.

e. In the case where cleaned tiles can also become dirty, the values of the previous grid points stored in the memory of the robot lose their importance, ie. since the values of the grid points can change with time, giving the best possible result becomes very difficult. For example, the tile which has just been cleaned may become very dirty within a few moves, but the robot will have its dirt value stored as 0. So what we should do is use the simple greedy mechanism istead of the part C mechanism.


E.
a. If the robot has to be designed specific to some given environment then the colour n texture of the floor can be provided to the robot as prior information, and with this, the robot can make out where the floor is, in the given image. We can do the similar thing for walls also. However, this method becomes really hard to implement when it comes to designing a robot which works in different environments, a lot of data will have to be given as prior information. The robot can detect the walls by using a simple edge detection algorithm.

b. There are 3 different types of dirt in the given picture, white powder, white paper pieces, and tea cup. If we take a 5X5 grid on the picture, with starting positon as (0,0). Then white powder can be mapped to (0,0) and (0,1). Pieces of paper can be mapped to (1,3) and (2,3). Tea cup can be mapped to (3,2).

c. No, the values between 0 and 1 will not be enough to map the dirt in each grid, since the dirt is of different types, we need to add other indicators as well. For example, if there are paper pieces on one tile and leaves on the other, we cannot compare that which is dirtier. There might even be usefull things on the floor, we need to take care that the robot can distinguish them from dirt, thus values from 0 to 1 are not enough.

d. We can consider that the robot only takes pics of the floor looking downwards, then the position of the robot in the 5X5 grid (where the first block is (0,0)) can be given by (4,2)since the robot will be at the centre in terms of columns and at the starting in terms of rows.