COMP 4640 Intelligent & Interactive Systems
Programs Supporting Model - Based Reflex Agents
November 2008 Dr. Cheryl Seals
Simple reflex agents
Programs that support Model - based Reflex Agents
Simple reflex agents select precepts based on the current percept ignoring the rest of the precept history Example: Beetle
Model-based reflex agents
Programs that support Model - based Reflex Agents
Most systems are based on conditionaction rules
(i.e. situation-action rules, productions, or if-then rules) (e.g. If car-in-front is braking then initiate-braking p46)
Model-Based Reflex Agents
Most effective way to keep track of the part of the world it cant see now. Maintain some internal state that depends on percept history and thereby reflects at least some of the unobserved aspects of the current state (e.g. using some type of variable).
Production Based Systems
The production rule paradigm originated in the field of AI with the expert systems rule languages such as OPS5 (Brownston et al. 1985)
condition action
An inference engine cycles through all the rules in the system matching the condition parts of the rules with data in working memory. Of all the rules that match (the candidate set), one is selected using some conflict resolution policy and this selected rule is fired, that is, its action part is executed. The action part may modify the working memory, possibly according to the matched data and the cycle continues until no more rules match. Rule based
Rules have special ops: Fire, which causes a rule to be triggered Enable, which causes a rule to be activated Disable, which causes a rule to be deactivated
Conflict resolution
Break ties with Specification, Sequencing, Meta rules
Production Based Systems
CLIPS (CLanguage Integration Production System)
Production system developed at NASAs Johnson space center. Written in ANSI C instead of LISP CLIPS implements standard forward-chaining patternmatching algorithm CLIPS knowledge representation similar to OPS5 and ART systems. Constructs simple string fact assertion & retraction Templates If-then rules (productions) Objects and instances
NASA uses clips in the following projects
Intelligent computer aided crew training, weather forecasting, shuttle space planning, shuttle diagnostics, Mission Control Center (telemetry data analysis and diagnostics), flight assistance and control
ART commercial expert system has many of the same features as CLIPS
Agent Based Systems
Systems to investigate Stagecast CreatorTM (www.stagecast.com) AgentsheetsTM (www.agentsheets.com)
End User Programming with agents Stagecast Study Report:
We are attempting to create a cross-generational web based learning community for middle school students, teachers, and seniors. Learning community will design, construct, and discuss simulations of community issues. Summary of results of formative evaluation with students creating simulation projects.
Proceedings of IEEE Visual Languages 2001, Rosson, Seals 2001; CHI 2001; DIS 2002; NSF Research: NSF ITR 0091102.
Stagecast Creator
Based on a movie metaphor Programming is facilitated by macro recorder to allow programming by demonstration Behaviors are represented as a set of as a set of productions or if-then rules
Procedure
Participants: 10 middle school students Background survey Performed in usability testing lab study with think aloud protocol Recorded critical incidents Captured video, audio, and screen
Subjective questionnaire, knowledge survey, retrospective interview
Visual Agent Programming
Spatial context and visual appearance are required elements in a rules precondition Correct position and appearance are preconditions for rules
Characters may have many instantiations
If Precondition is satisfied, Then rule is fired.
Observations and Results
Duration 30-55 minutes Activity I Duration 34-47 minutes Activity II Most students were successful in modifying simulations and adding new characters. Usability satisfaction
Easy and fun to use Would like to use it in their classes But needed more exposure to feel confident No problems with drawing tools Problems with tools for rule creation
Stagecast Usability Problems
Issue
Directing input to the wrong window Confused between rules and rule-actions lists Select wrong icon Inability to find rules or other content in window Misunderstand spotlight and concept of stretching it
Likely Cause
Too many similar-looking windows Lists that look similar but have different meanings Multiple similar icons Non-traditional method of scrolling Spotlight metaphor is not obvious or intuitive
Visual Programming Challenges
Practical metaphors for icons Bigger Icons Fewer layers of scaffolding Relation between internal variables and visual state of the simulation. Role of visual context in rules
Rules must match exact visual context, most PBD system make rules to specific to be reused
End User Programming with agents AgentSheets Study Report:
AgentSheets is a production based visual programming language where end users create with direct manipulation techniques
Reports a study of teachers learning to build educational simulations as curricula aids. Summary of results of formative evaluation to design agent based production system for end user creation of educational simulations. Proceedings of IEEE Visual Languages 2002, Seals 2002.
Example Rule
- left-hand specifies a before state - right-hand specifies one or more actions to take if state is confirmed
- multiple rules are tested in order, first match fires
Empirical Study Results
Need robust drawing tools Objects should be important, not their spatial location Flexible object size Support for import of objects Allow incremental testing Increase the level of usability for novice programmers Platform independent implementation
