artificial intelligence and lisp #3 characterization of actions [chapter 5] representation language...

23
Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Upload: isabel-robinson

Post on 16-Dec-2015

215 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Artificial Intelligence and Lisp #3

Characterization of Actions [Chapter 5]

Representation Language [Chapter 2]

Lab Assignment 2

Page 2: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Part I: Working with action laws

[Read Lecture Notes Chapter 5 for this,but read Chapter 2 first]

Page 3: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Effect law for pour action

[Do .t [pour .fr .to]] = [if [and [H- .t (substance-in: .to) empty] (not [H- .t (substance-in: .fr) Empty ])] [coact [H! .t (substance-in: .to) (cv (substance-in: .fr)) ] [H! .t (substance-in: .fr) Empty ]] [soact] ]]

Page 4: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Notation on previous slide

[H- time feature val] is a logical expression which is true if the value of feature at time-1 is val

[H time feature val] is a logical expression which is true if the value of feature at time is val

[H! time feature val] is a low-level “action” that assigns val as the value of feature at time

Page 5: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Use of effect law in simulator Check whether precondition is satisfied If so, execute the 'then' clause in the

environment. A composite action (coact) always takes one

timestep. A sequence (soact) takes one timestep per component.

If precondition is not satisfied, then execute the 'else' clause.

Report success or failure, failed precondition, and culprit of the failure (so that the agent has a chance to fix the problem)

Page 6: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Precondition for pour action

[Do .t [pour .fr .to]] = [if [and [H- .t (substance-in: .to) empty] (not [H- .t (substance-in: .fr) empty ])] [coact [H! .t (substance-in: .to) (cv (substance-in: .fr)) ] [H! .t (substance-in: .fr) empty ]] [soact] ]]

Failed precondition as reported by simulator: [H- now (substance-in: g2) empty]

Page 7: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Operations done on the precondition

Evaluate precondition. If failed: Instantiate variables (most of them) Insert now for current time in simulated world Identify which (one or more) of the conjuncts

has the value false Return both entire instantiated precondition,

and failed conjuncts More complicated operation if the precondition

is more complicated

Page 8: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Agent's response to failed precondition: alternatives

Recall: [H- now (substance-in: g2) empty] Simplest: Look up reaction in table - but based on what? Associate reaction(s) with pattern: [H- now (substance-in: .g) empty] [H- now (substance-in: .g) .c] Check postconditions of all available actions to find one

with desired effect: dispose [H! .t (substance-in: .e) empty] Find sequence of actions (a plan) whose combined effect

will be the desired one.

Page 9: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Operations on action expressions

Consist of expressions for precondition and for effect which are treated separately

Evaluation of precondition Partial evaluation of precondition Effect is ~ postcondition Match partially evaluated precondition with

postcondition/effect of another action Send expression to another agent Save expression for later reuse (learning)

Page 10: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Part II: Representation formalism[Read Lecture Notes, Chapter 2 for parts II-IV]

Requirements: Adequate expressivity: facts, scripts, action

laws Facilitate processing in computer Readability for humans Conciseness - avoid multiple representations Use available standards both from mathematics

and logic, and from computer world

Page 11: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

KRF - atomic elements

Atomic entities, e.g. Plato Strings, e.g. ”Ursus ursus” Numbers, e.g. 451

Page 12: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

KRF - composite elements

Composite entities, e.g. (age-of: Plato) Sets, e.g. {Plato Socrates} Sequences, e.g. <Caesar Augustus Nero> Records, e.g. [dispose :e glass4] Maplets, e.g. [: 45 empty] Forms, e.g. (cv (age-of: Plato)) or (+ 2 (cv (age-of: Plato)))

Page 13: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Composite entities vs forms

Composite entities, e.g. (age-of: Plato) Forms, e.g. (cv (age-of: Plato)) or (+ 2 (cv (age-of: Plato))) Composite entities can have attributes and

values Forms are replaced by their value during

evaluation - composite entities are not

Page 14: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

KRF - records

Records, e.g. [dispose :e glass4] More generally: [oper arg1 arg2... :tag1 val1 :tag2 val2 ... ] Alternative, equivalent: [oper :tag1 val1 :tag2 val2 ... ^ arg1 arg2... ]

Page 15: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

KRF - mappings

Maplets, e.g. [: 45 empty] Used to form mappings, e.g. {[: 21 water] [: 28 empty] [: 35 beer] [: 45 empty]}

Page 16: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Part III: Generic Scripting Language

Used as a basis for a range of scripting languages:

Agent Scripting Language Document Scripting Language (for texts and

web pages) Operations Scripting Language ...

Page 17: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Scripting Expressions

Action expression, record, [verb a1 a2...] Action composition using if, soact, etc. Feature, comp.ent., (feaop: arg1 ...) Term, form, (function arg1 arg2 ...) Query, form, is a kind of term Atomic propos., record, [pred a1 a2 ...] Composite propos., form, (and p1 p2 ...)

and similarly for or, not, etc.

Page 18: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Part IV: Entity-files

Entity-files are used in Leonardo for persistent storage of information between sessions

In principle, each entityfile is a two-level mapping:

{[: ent-1 {[: a1 val-1-1] [: a2 val-1-2] ...}] [: ent-2 {[: a1 val-2-1] [: a2 val-2-2] ...}] ... ... }

Page 19: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Practical representation of entityfiles------------------------------------- entity-1

[: type sometype][: a1 “value of attribute a1 for entity-1”][: a2 (symbfun: a5)][: (attrfun: color) {red green}]

------------------------------------- entity-2

[: type sometype][: a1 “value of attribute a1 for entity-2”]

@CommentComments may be expressed in free text, or mayuse a specialized notations as selected by thedesigner. They are preserved in textual formuntil processed by an application program.

ooooooooooooooooooooooooooooooooooo

Page 20: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Entityfile facilities

Each entity in the file must have a value for the type attribute. That value must again be an entity in a file, and so on.

Entityfiles can be written and read back. Only attributes that are included in the attributes attribute of the type are written to the file

Entities not having a value for any of the attributes of their type, are not written to file

Notice the use of properties, e.g. Comment

Page 21: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Some additional aspects An entity of type section can be used for

partitioning an entityfile into parts separated by a =================== line

The in-categories attribute is always optional and should have a set of entities as value

The command sortfil myfile sorts the entities in a file in alphabetical order

The name of an entityfile is always the first entity in the file

Page 22: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

Lab 2: Defining a Zoo Miniworld

Milestone 1: static structure in the zoo world Milestone 2: defining actions in the zoo world What to do for the lab:

Page 23: Artificial Intelligence and Lisp #3 Characterization of Actions [Chapter 5] Representation Language [Chapter 2] Lab Assignment 2

What to do for the lab(your “script”)

Read lab instructions on the course webpage: page on labserver access and lab 2 definition

Start a session with your Leonardo agent Download lab materials from labserver using

commands getlabs and getlab lab2a-kb Do the work as specified Make a local check using check-lab command Upload results using upload-lab command Await final approval