aaai 2014 spring - learning task management of an aircraft approach system
DESCRIPTION
Validating models of airspace operations is a particular challenge. These models are often aimed at finding and exploring safety violations, and aim to be accurate representations of real-world behavior. However, the rules governing the behavior are quite complex: nonlinear physics, operational modes, human behavior, and stochastic environmental concerns all determine the responses of the system. %In order to quantify uncertainty in the model (and by extension, risk in the real world), one recently successful methodology has been to develop a response surface replacement for the original model, and to learn the behavior of the system from the response surface. In this paper, we present a study on aircraft runway approaches as modeled in Georgia Tech's Work Models that Compute (WMC) simulation. We use a new learner, Genetic-Active Learning for Search-Based Software Engineering (GALE) to discover the Pareto frontiers defined by cognitive structures. These cognitive structures organize the prioritization and assignment of tasks of each pilot during approaches. We discuss the benefits of our approach, and also discuss future work necessary to enable uncertainty quantification.TRANSCRIPT
National Aeronautics and Space Administration
www.nasa.gov
Learning the Task Management Space of an Aircraft Approach Model
Dr. Misty D. DaviesResearch Computer EngineerNASA Ames Research Center
AAAI Conference, Spring 2014March 24, 2014
Joseph H. KrallPh.D. Candidate, Comp. Sci.West Virginia University
Dr. Tim MenziesProf., Comp. Sci.West Virginia University
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 2
• CDA:– Improved efficiency– Reduced Emissions– Less Noise for the city
Motivation Solutions Results Closings
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 3
• Modeling CDA with WMC:– Want to find the response surface– What decisions optimize the objectives?
Motivation Solutions Results Closings
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 4
• CDA is a “Wicked” model.– No stopping rule. Run out of time or money.– No right or wrong. Only "better," "worse,"– Every wicked problem is essentially unique and novel.– All solutions are "one-shot operations."– No alternative solutions.
• What possible solutions exist for “wicked”?– SBSE Tools?
Motivation Solutions Results Closings
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 5
• We applied SBSE practices to explore CDA– Using tools called MOEAs– NSGA-II, SPEA2, and GALE
Motivation Solutions Results Closings
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 6
• NSGA-II and SPEA2:– Standard MOEA– Random Mutation– Explore thousands of options
• And GALE:– Active & Spectral Learner– Directionally Guided Mutation– Explore very few (20-50) options
Motivation Solutions Results Closings
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 7
• Summary Results:– Relative Percentages of Baseline– Stars = Significant winners
Motivation Solutions Results Closings
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 8
• Some Current Work: Huge Studies with GALE/CDA
• Different modes of CDA. Different HTM– 1. Fixed HTM (8 levels)– 2. Fixed HTM and Exclude Opportunistic Mode
• 20 repeats * 16 modes = 320 Runs of GALE– 83 hours of GALE– Parallel Gains:
• 12 hours with 8 concurrent Processes
Motivation Solutions ClosingsResults
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 9
• Cognitive Control Mode– Decisions require OPP and HIGH to mitigate low HTM
Motivation Solutions ClosingsResults
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 10
Motivation Solutions ClosingsResults
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 11
• Excluding OPP normalized most decisions– Sanity Check. But objective scores are worse.
– But scores were worse overall
Motivation Solutions ClosingsResults
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 12
Motivation Solutions ClosingsResults
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 13
• GALE can find solutions to CDA– Often better than those found with NSGA-II/SPEA2
• GALE makes huge studies possible– 83 hrs vs 70 (predicted) days with NSGA-II/SPEA2
• GALE finds some sanity– And can explain:– Effect of OPP vs HTM
Motivation Solutions Results Closings
04/12/2023 Learning the Task Management Space of an Aircraft Approach Model 14
• Questions?
~~~Successful Approach! Err… Closing.~~~ Closings