sterling somers : [email protected]@sterlingsomers.com dr. robert west :...
TRANSCRIPT
Sterling Somers : [email protected]. Robert West : [email protected]
Both @Institute of Cognitive ScienceCarleton UniversityOttawa, ON, Canada
Macro Cognition : Using SGOMS to Pilot a Flight Simulator
• SGOMS is an extension of GOMS to deal with interruptions, teamwork, and re-planning
• Adds a control structure above unit tasks – planning unit• SGOMS in ACT-R
• A particular way to program ACT-R to do complex tasks• Adds more buffers to ACT-R
• Planning unit buffer• Unit task buffer• Method buffer
• These can be thought of collectively as the goal module• To test SGOMS we need to show that it works across tasks
SGOMS
Sandwiches
Different SGOMS models in the lab
Why Aviation?
1. Real World2. Dynamic3. Both Planned and Reactive Behaviour4. Replanning5. Expertise6. Continuous / Feedback
InterfaceSimulator
• FlightGear• Free• Open Source (hack simulation loop?)
Custom Python API• Property tree via UDP/HTTP/Telnet• http/telnet simultaneously (read/write)
• State / Control of aircraft
Problem• Real time VS ACT-R time (for now…)
Simple Vision• Python bindings to OpenCV
• Very efficient• Redundancy check possible
VisionTop Down:
• SOS VisionVision Module:
• When not busy• Request from environment• Motion Detection
Python bindings OpenCV• Screen shot VS centre of screen comparison / frame
Motion Detection
Screen Shot at frame 1 Screen Shot at frame 2(arrow indicates direction of movement)
Vision : Dynamic Salience / Context Dependence
Vision uses salience factor (0-1) to determine probability of seeing an environment item
• Salience of Horizontal and Vertical movement calculated according to size of movement (in pixels )
• is pixel movement threshold• Possible different for different contexts?
Operator PoolStart Engine Press Both Brakes Recall Heading Maintain Heading
To Full Throttle Set Trim Look Heading Rudder Left/Right
Motor MethodsTo Full Throttle:Continue increasing throttle until lever stops moving
Planning Unit: Prepare for take off
DM
Unit Task: Get Started
Press Both Brakes+
Start Engine+
Cue: Motor module: not busyAuditory Module: no engine noise
Unit Task: Do settings
Set Trim+
Planning Unit: To take off speed
DM
Unit Task: Get to speed
Release Brakes+
To Full Throttle+
Unit Task: Watch Speed
Read Air Speed+
+
Planning Unit: To take off speed
DM
Unit Task: Get to speed
Release Brakes+
To Full Throttle+
Unit Task: Watch Speed
Read Air Speed+
Vision Module
Look at [target]
Look out window
Top-down request
Opportunistic, when no top-down request
See movement+ Reasonable sized move
Interruption
Maintain Heading
Rudder, etc.+
Clear Buffer+(resulting in resume)
Interruption• Interruptions and re-planning take place on different levels• Case 1
– Bottom up scanning– Small– Expected– Context remains the same (still taking off)– Handle it within the unit task
• Case 2– Bottom up scanning– Large– Unexpected– Context is changed (abort take off)– New unit task needed
• SGOMS Hypothesis – there is a set processing structure used across tasks
Thank you.
Sterling Somers : [email protected]
Dr. Robert West : [email protected]
Both @Institute of Cognitive ScienceCarleton UniversityOttawa, ON, Canada