A simple vacuum cleaner agent is having two sensors. One sensor senses the current location of the agent (floor A or floor B). The other sensor senses the status of the current location (Dirty or Clean). It is assumed that the agent can perform any one of the following three actions in a single time unit:
Action R: Agent moves from location A to location B in a single step i.e. single time unit.
Action L: Agent moves from location B to location A in a single step i.e. single time unit.
Action S: Agent sucks the dirt from current floor in a single step i.e. single time unit.
Calculate the Performance Measure of this Vacuum Cleaner agent for each of the following four initial conditions by assuming that the agent operates for 100 time units:
i) [A, Clean], [B, Clean]
ii) [A, Dirt], [B, Clean]
iii) [A, Clean], [B, Dirt]
iv) [A, Dirt], [B, Dirt]
Following criteria is assumed for measuring the performance:
- In each time unit, only one action (either R or L or S) takes place by the agent.
- +2 mark is rewarded to the agent for its every action (either R or L or S) till two floors are clean.
- -1 mark is given to the agent for its every move (either R or L) from one location to other after both the floors are clean.
My understanding of the problem:
For part (i) [A,Clean],[B,Clean] can be followed by,
[A,Clean],[B,Clean],[A,Clean] - Right
[A,Clean],[B,Clean],[A,Dirt] - Suck
[A,Clean],[B,Clean],[B,Clean] - Left
[A,Clean],[B,Clean],[B,Dirt] - Suck
This gives us only 4 actions. So +2 for each?
Since there are 100 time units given do we consider 100 actions since each action takes a single time unit? Does that mean we have to make a percept sequence with 100 actions?
If I assume the vacuum cleaner is a rational agent which would perform only the right actions according to the situation, then would it ever get -1 mark?
I am unsure if I am going about this problem correctly. How do I calculate the performance? Do I need to make the state space tree? Please help.