• Keine Ergebnisse gefunden

Planning and Optimization D2. Abstractions: Additive Abstractions Gabriele R¨oger and Thomas Keller

N/A
N/A
Protected

Academic year: 2022

Aktie "Planning and Optimization D2. Abstractions: Additive Abstractions Gabriele R¨oger and Thomas Keller"

Copied!
7
0
0

Wird geladen.... (Jetzt Volltext ansehen)

Volltext

(1)

Planning and Optimization

D2. Abstractions: Additive Abstractions

Gabriele R¨ oger and Thomas Keller

Universit¨ at Basel

October 29, 2018

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 1 / 27

Planning and Optimization

October 29, 2018 — D2. Abstractions: Additive Abstractions

D2.1 Multiple Abstractions D2.2 Additivity

D2.3 Outlook D2.4 Summary

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 2 / 27

Content of this Course

Planning

Classical

Tasks Progression/

Regression Complexity Heuristics

Probabilistic

MDPs Uninformed Search

Heuristic Search Monte-Carlo

Methods

Content of this Course: Heuristics

Heuristics

Delete Relaxation

Abstraction

Abstractions in General

Pattern Databases

Merge &

Shrink Landmarks

Potential Heuristics

Cost Partitioning

(2)

D2. Abstractions: Additive Abstractions Multiple Abstractions

D2.1 Multiple Abstractions

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 5 / 27

D2. Abstractions: Additive Abstractions Multiple Abstractions

Multiple Abstractions

I One important practical question is how to come up with a suitable abstraction mapping α.

I Indeed, there is usually a huge number of possibilities, and it is important to pick good abstractions

(i.e., ones that lead to informative heuristics).

I However, it is generally not necessary to commit to a single abstraction.

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 6 / 27

D2. Abstractions: Additive Abstractions Multiple Abstractions

Combining Multiple Abstractions

Maximizing several abstractions:

I Each abstraction mapping gives rise to an admissible heuristic.

I By computing the maximum of several admissible heuristics, we obtain another admissible heuristic which dominates the component heuristics.

I Thus, we can always compute several abstractions and maximize over the individual abstract goal distances.

Adding several abstractions:

I In some cases, we can even compute the sum of individual estimates and still stay admissible.

I Summation often leads to much higher estimates than maximization, so it is important to understand

D2. Abstractions: Additive Abstractions Multiple Abstractions

Adding Several Abstractions: Example (1)

LRR LLL LLR

LRL

ALR

ALL

BLL

BRL

ARL

ARR

BRR

BLR

RRR RRL

RLR

RLL

h (LRR) = 4

(3)

D2. Abstractions: Additive Abstractions Multiple Abstractions

Adding Several Abstractions: Example (2)

LRR LLR

LLL

LRL

LLR

LRL LLL

ALR ARL

ALL ARR

BLL

BRL BRR

BLR

ALR ARL

ARR ALL

BLL BRR

BLR BRL

RRR RRL

RLR

RLL

RLL

RRL

RLR RRR

h α

1

(LRR) = 3

LRR LLR

LLL

LRL ALR

ALL

BLL

BRL

LLR

LLL

LRL ALR

ALL

BLL

BRL

ARL

ARR

BLR BRR

RRR RRL

RLR RLL

ARL

ARR

BLR BRR

RRR RRL

RLR RLL

h α

2

(LRR) = 2

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 9 / 27

D2. Abstractions: Additive Abstractions Multiple Abstractions

Adding Several Abstractions: Example (3)

LRR LLR

LLL

LRL

LLR

LRL LLL

ALR ARL

ALL ARR

BLL

BRL BRR

BLR

ALR ARL

ARR ALL

BLL BRR

BLR BRL

RRR RRL

RLR RLL

RLL RRL

RLR RRR

h α

1

(LRR) = 3

LRR LLR

LLL

LRL ALR

ALL

BLL

BRL

LRR

LLR

LLL

LRL ALR

ALL

BLL

BRL

ARL

ARR

BLR BRR

RRR RRL

RLR RLL

ARL

ARR

BLR BRR

RRR RRL

RLR RLL

h α

2

(LRR) = 1

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 10 / 27

D2. Abstractions: Additive Abstractions Additivity

D2.2 Additivity

D2. Abstractions: Additive Abstractions Additivity

Orthogonality of Abstractions

Definition (Orthogonal)

Let α 1 and α 2 be abstractions of transition system T .

We say that α 1 and α 2 are orthogonal if for all transitions s − → ` t

of T , we have α i (s ) = α i (t) for at least one i ∈ {1, 2}.

(4)

D2. Abstractions: Additive Abstractions Additivity

Affecting Transition Labels

Definition (Affecting Transition Labels)

Let T be a transition system, and let ` be one of its labels.

We say that ` affects T if T has a transition s − → ` t with s 6= t.

Theorem (Affecting Labels vs. Orthogonality)

Let α 1 and α 2 be abstractions of transition system T . If no label of T affects both T α

1

and T α

2

,

then α 1 and α 2 are orthogonal.

(Easy proof omitted.)

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 13 / 27

D2. Abstractions: Additive Abstractions Additivity

Orthogonality and Additivity

Theorem (Additivity for Orthogonal Abstractions)

Let h α

1

, . . . , h α

n

be abstraction heuristics of the same transition system such that α i and α j are orthogonal for all i 6= j .

Then P n

i=1 h α

i

is a safe, goal-aware, admissible and consistent heuristic for Π.

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 14 / 27

D2. Abstractions: Additive Abstractions Additivity

Orthogonality and Additivity: Example (1)

LRR LLR

LLL

LRL

LLR

LRL LLL

ALR ARL

ALL ARR

BLL

BRL BRR

BLR

ALR ARL

ARR ALL

BLL BRR

BLR BRL

RRR RRL

RLR

RLL

RLL

RRL

RLR RRR

h α

1

(LRR) = 3

LRR LLR

LLL

LRL ALR

ALL

BLL

LLR

LLL

LRL ALR

ALL

BLL

ARL

ARR

BRR RRR RRL

RLR RLL

ARL

ARR

BRR RRR RRL

RLR RLL

h α

2

(LRR) = 2

D2. Abstractions: Additive Abstractions Additivity

Orthogonality and Additivity: Example (2)

LRR LLR

LLL

LRL

LLR

LRL LLL

ALR ARL

ALL ARR

BLL

BRL BRR

BLR

ALR ARL

ARR ALL

BLL BRR

BLR BRL

RRR RRL

RLR RLL

RLL RRL

RLR RRR

h α

1

(LRR) = 3

LRR LLR

LLL

LRL ALR

ALL

BLL

LRR

LLR

LLL

LRL ALR

ALL

BLL

ARL

ARR

BRR RRR RRL

RLR RLL

ARL

ARR

BRR RRR RRL

RLR RLL

h α

2

(LRR) = 1

(5)

D2. Abstractions: Additive Abstractions Additivity

Orthogonality and Additivity: Proof (1)

Proof.

We prove goal-awareness and consistency;

the other properties follow from these two.

Let T = hS , L, c , T , s 0 , S ? i be the concrete transition system.

Let h = P n i=1 h α

i

.

Goal-awareness: For goal states s ∈ S ? , h(s ) = P n

i=1 h α

i

(s ) = P n

i=1 0 = 0 because all individual

abstraction heuristics are goal-aware. . . .

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 17 / 27

D2. Abstractions: Additive Abstractions Additivity

Orthogonality and Additivity: Proof (2)

Proof (continued).

Consistency: Let s − → o t ∈ T . We must prove h(s) ≤ c (o) + h(t).

Because the abstractions are orthogonal, α i (s) 6= α i (t) for at most one i ∈ {1, . . . , n}.

Case 1: α i (s ) = α i (t) for all i ∈ {1, . . . , n}.

Then h(s ) = P n

i=1 h α

i

(s)

= P n

i=1 h T

αi

i (s ))

= P n

i=1 h T

αi

i (t))

= P n

i=1 h α

i

(t)

= h(t) ≤ c (o) + h(t).

. . .

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 18 / 27

D2. Abstractions: Additive Abstractions Additivity

Orthogonality and Additivity: Proof (3)

Proof (continued).

Case 2: α i (s) 6= α i (t) for exactly one i ∈ {1, . . . , n}.

Let k ∈ {1, . . . , n} such that α k (s ) 6= α k (t).

Then h(s) = P n

i=1 h α

i

(s)

= P

i∈{1,...,n}\{k} h T

αi

i (s)) + h α

k

(s)

≤ P

i∈{1,...,n}\{k} h T

αi

i (t)) + c (o) + h α

k

(t)

= c (o) + P n

i=1 h α

i

(t)

= c (o) + h(t),

where the inequality holds because α i (s ) = α i (t) for all i 6= k and h α

k

is consistent.

D2. Abstractions: Additive Abstractions Outlook

D2.3 Outlook

(6)

D2. Abstractions: Additive Abstractions Outlook

Using Abstraction Heuristics in Practice

In practice, there are conflicting goals for abstractions:

I we want to obtain an informative heuristic, but

I want to keep its representation small.

Abstractions have small representations if

I there are few abstract states and

I there is a succinct encoding for α.

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 21 / 27

D2. Abstractions: Additive Abstractions Outlook

Counterexample: One-State Abstraction

LRR

LLR

LLL

LRL

ALR

ALL

BLL

BRL

ARL

ARR

BRR

BLR

RRR RRL

RLR LRR RLL

LLR

LLL

LRL

ALR

ALL

BLL

BRL

ARL

ARR

BRR

BLR

RRR RRL

RLR

RLL

One-state abstraction: α(s ) := const.

+ very few abstract states and succinct encoding for α

− completely uninformative heuristic

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 22 / 27

D2. Abstractions: Additive Abstractions Outlook

Counterexample: Identity Abstraction

LRR LLL LLR

LRL

ALR

ALL

BLL

BRL

ARL

ARR

BRR

BLR

RRR RRL

RLR

RLL

Identity abstraction: α(s ) := s .

D2. Abstractions: Additive Abstractions Outlook

Counterexample: Perfect Abstraction

LRR

LLR

LLL

LRL LLR

LRL LLL

ALR

ALL

BLL

BRL ALR

BRL ALL

BLL

ARL

ARR

BRR

BLR ARL

BLR ARR

BRR

RRR RRL

RLR

RLL RLL RRL

RLR RRR

Perfect abstraction: α(s ) := h (s ).

(7)

D2. Abstractions: Additive Abstractions Outlook

Automatically Deriving Good Abstraction Heuristics

Abstraction Heuristics for Planning: Main Research Problem Automatically derive effective abstraction heuristics

for planning tasks.

we will study two state-of-the-art approaches in Chapters D3–D8

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 25 / 27

D2. Abstractions: Additive Abstractions Summary

D2.4 Summary

G. R¨oger, T. Keller (Universit¨at Basel) Planning and Optimization October 29, 2018 26 / 27

D2. Abstractions: Additive Abstractions Summary

Summary

I Often, multiple abstractions are used.

They can always be maximized admissibly.

I Adding abstraction heuristics is not always admissible.

When it is, it leads to a stronger heuristic than maximizing.

I Abstraction heuristics from orthogonal abstractions can be added without losing admissibility or consistency.

I One sufficient condition for orthogonality is that all abstractions are affected by disjoint sets of labels.

I Practically useful abstractions are those which give informative heuristics, yet have a small representation.

I Coming up with good abstractions automatically

is the main research challenge when applying

abstraction heuristics in planning.

Referenzen

ÄHNLICHE DOKUMENTE

Abstractions: Formal Definition and Heuristics Homomorphisms and Isomorphisms. D1.2 Homomorphisms

Abstractions α induce abstraction heuristics h α : h α (s ) is the goal distance of α(s ) in the abstract transition system. Abstraction heuristics are safe, goal-aware, admissible

Coming up with good abstractions automatically is the main research challenge when applying abstraction heuristics

D3.1 Projections and Pattern Database Heuristics D3.2 Implementing PDBs: Precomputation.. D3.3 Implementing PDBs: Lookup

Formally, pattern database heuristics are abstraction heuristics induced by a particular class of abstractions called projections.

I Hence we now look at the transition systems for atomic projections of our example task, including transition labels. I We abbreviate operator names as in

Hence we now look at the transition systems for atomic projections of our example task, including transition labels. We abbreviate operator names as in

The objective value of an integer program that minimizes this cost subject to the flow constraints is a lower bound on the plan cost (i.e., an admissible heuristic estimate)..