AI代写|6CCS3AIP Artificial Intelligence Planning


a. Consider a temporal PDDL planning problem, which models a robot
moving between two rooms. It is assumed that there is some remote
control that the robot can press in order to open the door and that it
will then remain open for 2 time units: notice that the open door action
deletes its own precondition and can only ever be applied once.

(:durative-action open-door
:parameters (?d – door)
:duration (= ?duration 2)
:condition (and
(at start (can-open ?d))
:effect (and
(at start (not (can-open ?d)))
(at start (open ?d))
(at end (not (open ?d)))
(:durative-action move
:parameters (?rob – robot ?from ?to – room ?d – door)
:duration (= ?duration 8)
:condition (and
(at start (at ?rob ?from))
(at end (open ?d))
(over all (door-between ?from ?to ?d))
:effect (and
(at start (not (at ?rob ?from)))
(at end (at ?rob ?to))

Initial State: (can-open door1) (at rob room1) (door-between
room1 room2 door1)

Goal:(at robot room2)

i. Show how a decision epoch planner would proceed to solve this prob
lem from the initial state, by showing the states explored (including
the propositions that are true, the timestamps of the states and the
event queue). Can the planner solve the problem? If so, show how
search proceeds to reach the goal; if not show the states that appear
in search until the planner gets stuck and explain why the problem is
not solvable from this point.
[6 marks]

ii. The search in CRIKEY3 generates plans that are propositionally sound,
and then uses an STN to check whether the plan is temporally sound.
Explain why the following plan is propositionally sound (despite not
being temporally sound).

(open-door door1) start
(move rob room1 room2 door1) start
(move rob room1 room2 door1) end
(open-door door1) start
[3 marks]

iii. Draw the simple temporal network (STN) that CRIKEY 3 would
produce for the plan in part ii (you may use the first letter of each
action and the symbols ‘ and a (for start and end) as abbreviations
in your diagram if you wish).
[4 marks]

iv. Show how you can deduce from your STN that this plan is not tem
porally valid.
[2 marks]

b. This question is about preferences. Consider the following planning
problem which models the daily activities of a cat.

(:action eat-tuna (:action eat-cat-food
:parameters () :parameters ()
:precondition (and :precondition (and
(been_cute) (in_house)
(in_house) )
:effect (and :effect (and
(had_tuna) (had_dinner)
(had_dinner) ))
(:action take-nap (:action play-with-human
:parameters () :parameters ()
:precondition (and :precondition (and
(in_house) (in_house)
) )
:effect (and :effect (and
(slept) (been_cute)
)) ))

Initial State: (in_house)

In an attempt to model the desires of a cat, that it must take a nap
before dinner and a nap after dinner, a human has specified the following
pair of preferences:

• (sometime-before had_dinner slept)
• (sometime-after had_dinner slept)

i. Sketch the automaton for the preference (sometime-after had-dinner
[3 marks]