A.I. PLANNING
A.I. PLANNING
Amanda Coles
Copyright By PowCoder代写 加微信 powcoder
“You know what the difference is between a dream and a goal?… A plan.” – Jodi Picoult
Still a good thing to do first
Domain Independent
“In any non-trivial situation, there is a choice about what to do next…”
Planning – The Problem
In planning problems we have:
an initial state, I
(at mydvd amazon)
(at truck amazon)
(at driver home)
(path home amazon)
(link amazon london)
(link london myhouse)
Closed world assumption
Planning – The Problem
In planning problems we have:
an initial state, I
(at mydvd myhouse)
Can be several facts, a conjunction of facts
Not a Goal state: goal is a partial specification of a state
Planning – The Problem
In planning problems we have:
an initial state, I
some actions, A, defined according to a domain
(load mydvd truck depot)
(walk driver home amazon)
(board-truck driver truck amazon)
(drive-truck driver amazon london)
Planning – The Problem
In planning problems we have:
an initial state, I
some actions, A, defined according to a domain
these have preconditions and effects
(:action load
(:parameters (?d – item ?t – truck ?p – place)
(:precondition (and (at ?t ?p) (at ?d ?p)))
(:effect (and (not (at ?d ?p))
(in ?d ?t))
Note at is deleted here: purpose of fact is a precondition, not just to model the world
Also note that this is a lifted action
Planning – The Problem
In planning problems we have:
an initial state, I
some actions, A, defined in a domain
A planner takes these and gives us a plan.
An ordered sequence of actions from A that will turn I into G
what to do; and when to do it.
Performs search to consider the different possible plans available, until a plan from I to G is found.
Planning More Formally
A classical (STRIPS) planning problem is a tuple
P is a finite set of propositions;
A is a set of actions each of which has:
A precondition, a set of propositions that must be true in order for A to be executed;
Add effects: a set of propositions added that become true upon execution of A;
Delete effects: A set of propositions that are deleted upon execution of A.
I a finite set of propositions representing the initial state
G the goal, a set of propositions that must be true for a state to be considered a goal state.
(at mydvd amazon)
(at truck amazon)
(at driver home)
(path home amazon)
(link amazon london)
(link london myhouse)
(at mydvd myhouse)
Initial: (nostranger tolove), (knows rules)
Goal: (so do I)
Action Preconditions Effects
Never (full commitment)
(knows rules) (not (full commitment))
Gonna (knows rules) (not (knows rules))
(any other guy)
Give (any other guy) (not (any other guy))
(knows rules)
YouUp (nostranger tolove)
(any other guy) (not (nostranger tolove))
(full commitment)
Back to the Easier Problems…
How might one encode this in PDDL? Initial state: right door is open; standing in 1st room. Goal: standing in third room.
Right door open (knows rules)
In First room (nostranger tolove)
Goal of standing in right room (so do I)
Actions for buttons: Gonna and Give
To move, Never and YouUp
Intuition is a powerful tool.
Pressing a red button opens one door and shuts the other
Why should I care?
We know: the current state of the world: Rover at X, energy supply adequate, package at Y etc.
The desired state of the world: Science data, continued adequate supply, package at Z, clean house, sell goods etc.
What actions we can carry out (under which conditions): Sample soil, switch on power station, drive truck, clean floor, print page.
We want to know what to do and in what order to do it.
Warehouse Video 1:32
Animation @ 12:00
Example plan @ 26:17
程序代写 CS代考 加微信: powcoder QQ: 1823890830 Email: powcoder@163.com