1 / 27

Predictive Engineering Models Based on the EPIC Achitecture for a Multimodal High-Performance

Predictive Engineering Models Based on the EPIC Achitecture for a Multimodal High-Performance Human-Computer Interaction Task. Abstract Introduction The EPIC Architecture The Telephone Operator Task Modeling the Telephone Operator Task A Set of A Priori Models

cole
Download Presentation

Predictive Engineering Models Based on the EPIC Achitecture for a Multimodal High-Performance

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Predictive Engineering Models Based on the EPIC Achitecture for a Multimodal High-Performance Human-Computer Interaction Task

  2. Abstract Introduction The EPIC Architecture The Telephone Operator Task Modeling the Telephone Operator Task A Set of A Priori Models Comparison of the Models To Empirical Data Results Conclusion Contents

  3. Engineering model can replace to some extent expensive user-testing data. The EPIC models are a feasible approach to predicting performance in multimodal high-performance tasks. Abstract

  4. Introduction Background on CPM-GOMS - Based on MHP and Critical Path Method Generative Models of Procedures Using the EPIC Architecture - powerful and easier to apply than CPM-GOMS - Such as Aircraft cockpit task Can EPIC Models Predict Performance?

  5. The EPIC Architecture

  6. The EPIC Architecture Perceptual Processor Visual Processor Auditory Processor Cognitive Processor Production Rule and Cycle Time Working Memory Movement Preparation and Execution Manual Motor Processor Motor Processor Vocal Motor Processor Oculomotor Processor Comparion with the MHP Motor Processor

  7. Perceptual Processors (Visual) Simple “Pipe line” no time intergration effect Detection Timing : 50 ms. Shape information Timing : 100 ms Pattern recognition Timing : 250ms Fovea (1°) : CONTENS Parafovea (10°) : CRUDER Periphery (60°) : LOCATION

  8. Perceptual Processors (Auditory) Corresponding onset of the tone : 50msec Discriminate frequency : 250msec offset item : 50msec Item disappear : 4sec Speech input is a item for single word Recognize input word : 150msec Tag-chained sequence

  9. Cognitive Processor Parsimonious Production System (PPS) “<rule name>if<condition>then<actions>” Timing: 50 ms/cycle Parallel cognitive processor -> Multitask Control store/ Working memory No lost & No Limit W/M

  10. Motor Processors Preparation / Excution phase Features can be prepared in advance or re-used. Later execution is faster. Timing: 50 ms/feature preparation. 50 ms movement initiation delay.

  11. Motor Processors Manual Motor Processor - Peck movement style (100msec) - Fitts’ law Vocal Motor Processor - Each Utterance is a single symbol (100msec) Oculomotor Processors (voluntary & involuntary) - 4msec/deg Comparison with the MHP Motor Processor

  12. The Telephone Operator Task Task Description Rationale for Using the Bill To Call Type Internally and Externally Determined Events Strong Test of Models Require Internally Determined Events

  13. General Modeling Properities of EPIC Fixed 1. The connection and mechanisms of processors. 2. Most time parameters. 3. The feature structures and time parameters of motor processors. Free to vary 1.Task-specific production rule programming 2.Task-specific perceptual encoding type and times 3.The style of movements Often not determined by the task

  14. Modeling the Telephone Operator Task Choosing Task Strategies For Models : The need for modeling Policies Some Possible Policies for Overlapping Task Activities

  15. A set of a priori models Choice of Models - Best Optmized Model - Simpler Model Simplifications Based on the operator’s expertise • -. The operator are very well practiced • -. Eye movement can be made directly to the field • -. Before Key stroke, eye must be moved to them • -. Then their shape must be available in visual working memory • (STA-SPL-CLG, KP-SPL, DIGIT NUMBER KEYS) • . Only POS-RLS keyvisual acquisition required

  16. Hierarchical Models Method for goal: Do CCS Task Step 1. Wait for call arrival tone Step 2. Look at the CLG-Type field Step 3. Retain CLG-Type and Accomplish goal: Handle CLG Type Step 4. Press POS-RLS Step 5. Return with goal accomplished Selection rule set for goal: Handle CLG Type If CLG-Type is 0+ then Accomplish goal: Handle zero-plus-call If CLG-Type is 0 then Accomplish goal: Handle zero-call If CLG-Type is 1+ then Accomplish goal: Handle one-plus-call If CLG-Type is OVT then Accomplish goal: Handle overtime-call Return with goal accomplished Method for goal: Handle zero-plus-call Step 1. Retain payment is paid and call is station-to-station Step 2. Accomplish goal: Greet customer Step 2. Accomplish goal: Get customer request Step 5. Accomplish goal: Do call type verification Step 6. Return with goal accomplished Method for goal: Greet customer Step 1. Look at coin-pre field EXPERT OF NGOML METHOD

  17. NGOMSL VS EPIC EPIC (*MFG-Do-CCS IF ((GOAL DO CCS TASK) (NOT (WM Executing CCS Task)) (MOTOR OCULAR MODALITY FREE)) THEN ((SEND-TO-MOTOR OCULAR FIXATE FIXATION-POINT) (ADDDB (STEP Waiting for call)) (ADDDB (WM Executing CCS Task)))) (*Do-CCS*Wait-for-CAT IF ((GOAL DO CCS TASK)(STEP Waiting for call) (AUDITORY DETECTION EVENT ONSET)) THEN ((DELDB (STEP Waiting for call)) (ADDDB (STEP Look at CLG type)))) NGOMSL Method for goal: Do CCS Task Step 1. Wait for call arrival tone Step 2. Look at the CLG-Type field Step 3. Retain CLG-Type and Accomplish goal: Handle CLG Type Step 4. Press POS-RLS Step 5. Return with goal accomplished Selection rule set for goal: Handle CLG Type If CLG-Type is 0+ then Accomplish goal: Handle zero-plus-call If CLG-Type is 0 then Accomplish goal: Handle zero-call If CLG-Type is 1+ then Accomplish goal: Handle one-plus-call If CLG-Type is OVT then Accomplish goal: Handle overtime-call Return with goal accomplished

  18. Hierarchical Models Fully-Sequential Policy (*Get-Billing-Number*Look-at-STA-SPL-KEY IF ((GOAL Get billing number) (STEP Look at STA-SPL-CLG) (MOTOR OCULAR MODALITY FREE)) THEN ((SEND-TO-MOTOR OCULAR FIXATE STA-SPL-CLG-KEY) (DELDB (STEP Look at STA-SPL-CLG)) (ADDDB (STEP Press STA-SPL-CLG key)))) (*Get-Billing-Number*Press-STA-SPL-CLG IF ((GOAL Get billing number) (STEP Press STA-SPL-CLG key) (VISUAL ??? SHAPE STA-SPL-CLG-KEY) (MOTOR OCULAR MODALITY FREE) (MOTOR MANUAL MODALITY FREE)) THEN ((SEND-TO-MOTOR MANUAL PERFORM Peck STA-SPL-CLG-KEY) (DELDB (STEP Press STA-SPL-CLG key)) (ADDDB (STEP Press KP-SPL key)))) Motor-Parallel Policy (*Get-Billing-Number*Look-at-STA-SPL-KEY IF ((GOAL Get billing number) (STEP Look at STA-SPL-CLG) (MOTOR OCULAR PROCESSOR FREE)) THEN ((SEND-TO-MOTOR OCULAR FIXATE STA-SPL-CLG-KEY) (DELDB (STEP Look at STA-SPL-CLG)) (ADDDB (STEP Press STA-SPL-CLG key)))) (*Get-Billing-Number*Press-STA-SPL-CLG IF ((GOAL Get billing number) (STEP Press STA-SPL-CLG key) (VISUAL ??? SHAPE STA-SPL-CLG-KEY) (MOTOR MANUAL PROCESSOR FREE)) THEN ((SEND-TO-MOTOR MANUAL PERFORM Peck STA-SPL-CLG-KEY) (DELDB (STEP Press STA-SPL-CLG key)) (ADDDB (STEP Press KP-SPL key))))

  19. Hierarchical Models PREPARED MOTOR PARALLEL (*Handle-Zero-Plus-Call*PrepareSTA-SPL-CLG IF ((Goal Do Zero-Plus Task) (STEP Greet customer) (NOT (STEP PREPARE STA-SPL-CLG-KEY)) (NOT (WM PREPARED STA-SPL-CLG-KEY))) THEN ((ADDDB (STEP PREPARE STA-SPL-CLG-KEY)))) (*Prepare*STA-SPL-CLG IF ((STEP PREPARE STA-SPL-CLG-KEY) (MOTOR MANUAL PROCESSOR FREE)) THEN ((SEND-TO-MOTOR MANUAL PREPARE Peck STA-SPL-CLG-KEY) (ADDDB (WM PREPARED STA-SPL-CLG-KEY)) (DELDB (STEP PREPARE STA-SPL-CLG-KEY)))) PREMOVE PREPARED MOTOR PARALLEL (*Handle-Zero-Plus-Call*SetupSTA-SPL-CLG IF ((Goal Do Zero-Plus Task) (STEP Greet customer) (NOT (WM SETUP STA-SPL-CLG-KEY IN PROGRESS)) (NOT (WM PREPARED STA-SPL-CLG-KEY))) THEN ((ADDDB (WM SETUP STA-SPL-CLG-KEY IN PROGRESS)) (ADDDB (STEP PREMOVE STA-SPL-CLG-KEY)))) (*Premove*STA-SPL-CLG IF ((STEP PREMOVE STA-SPL-CLG-KEY) (MOTOR MANUAL PROCESSOR FREE)) THEN ((SEND-TO-MOTOR MANUAL PERFORM POSE STA-SPL-CLG-KEY))

  20. FLATTENED Models FLATTENED MOTOR PARALLEL (*Do-CCS*0+Look-at-Coin-Pre IF ((GOAL DO CCS TASK) (STEP Look at coin-pre) (VISUAL ?OBJECT SHAPE CLG-TYPE) (VISUAL ?OBJECT LABEL 0+) (MOTOR OCULAR PROCESSOR FREE)) THEN ((SEND-TO-MOTOR OCULAR FIXATE COIN-PRE-FIELD) (ADDDB (WM CLG type is 0+)) (ADDDB (WM Call is for station)) (ADDDB (WM Payment is paid)) (DELDB (STEP Look at coin-pre)) (ADDDB (STEP Decide coin-pre)))) HIERACHICAL Motor-Parallel (*Get-Billing-Number*Press-STA-SPL-CLG IF ((GOAL Get billing number) (STEP Press STA-SPL-CLG key) (VISUAL ??? SHAPE STA-SPL-CLG-KEY) (MOTOR MANUAL PROCESSOR FREE)) THEN ((SEND-TO-MOTOR MANUAL PERFORM Peck STA-SPL-CLG-KEY) (DELDB (STEP Press STA-SPL-CLG key)) (ADDDB (STEP Press KP-SPL key)))) (*Get-Billing-Number*Press-KP-SPL-key

  21. Comparison Of The Models To Empirical Data Observed and Predicted Times Details of the Videotaped Data Details of the Simulation Process A baseline model is KLM(Keystroke-Level Model) - Total Task Time = Workstation response time + customer and operator speaking time + Total time for keystroke (280ms each) + Homing operators for movement (400ms each)

  22. Results Prediction of Total Task Execution Times call arrival signal tone – POS-RLS key stroke (x) call arrival signal tone – START key stroke (o)

  23. Goodness of fit Measures • 83% variance in p<0.05 회귀분석이 의미 없음 • Average absolute error of prediction • Under predicted data

  24. Results Comparison of EPIC Models to CPM-GOMS EPIC generating usefully accurate predictions for the selected task instance EPIC could be readily applied to a larger number of task instances

  25. Results Implications of individual keystroke time predictions • Speech Recognition delay and input rate • - What controls the POS-RLS keystroke?

  26. EPIC models could predict total execution times with an accuracy enough. Model was easy to construct. (comparing with CPM-GOMS) At this point(1997), EPIC is a research system and not ready for routine use. Designers can conveniently apply engineering models based on EPIC. CONCLUSIONS

More Related