A Case for Theory-Based Research on Level of Automation and
Adaptive Automation
David B. Kaber
Department of Industrial Engineering
North Carolina State University
This work was supported by a grant through the National Aeronautics and Space Administration Langley Research Center. The opinions expressed are those of the author and do not necessarily reflect the views of NASA.
Thoughts on the “what”, “how” and “when” of automation:
“What?”
“How?”
“When?”Monitoring
Generating
Selecting
Implementing
Human
Human/Machine
Machine
Stages of Information Processing
Level of Automation (LOA)
Adapted from Endsley (1997).
TnT0
Adaptive Automation (AA)
Information Processing and Automation Design Theory:
Qualitative models relating stages of information processing to automation design.– Parasuraman et al. (2000) – Model of types and
levels of automation.
A model of types and levels of automation (from Parasuraman et al. (2000)).
Information Acquisition
Information Analysis
Decision-making
Action
Degree of Automation
High
Low
High
Low
High
Low
High
Low
Qualitative Models of Human-Automation Interaction (cont.):
Endsley & Kaber (1999):– Four-stage model of information processing.– Used to define levels of automation or function
allocation schemes.
Monitoring Generating Selecting Implementing
Human
Human/Machine
Machine
Level of Automation
Human
Human/Machine
Machine
Human
Human/Machine
Machine
Human
Human/Machine
MachineAdapted from Endsley and Kaber’s (1999) taxonomy of levels of automation.
Taxonomy of Levels of Automation:
Endsley & Kaber (1999):– 10 discrete levels of automation (LOAs) in complex
systems control. Roles
Level of Automation MONITORING GENERATING SELECTING IMPLEMENTING
(1) MANUAL CONTROL Human Human Human Human
(2) ACTION SUPPORT Human/Computer Human Human Human/Computer
(3) BATCH PROCESSING Human/Computer Human Human Computer
(4) SHARED CONTROL Human/Computer Human/Computer Human Human/Computer
(5) DECISION SUPPORT Human/Computer Human/Computer Human Computer
(6) BLENDED DECISION MAKING Human/Computer Human/Computer Human/Computer Computer
(7) RIGID SYSTEM Human/Computer Computer Human Computer
(8) AUTOMATED DECISION MAKING Human/Computer Human/Computer Computer Computer
(9) SUPERVISORY CONTROL Human/Computer Computer Computer Computer
(10) FULL AUTOMATION Computer Computer Computer Computer
Bases for Qualitative Models:
Common models of HUMAN information processing (HIP).– Sanders & McCormick (1993) – Multi-stage model
identifying functions of humans and machines.
Adapted from Sanders and McCormick’s (1993) model of human and machine functions in human-machine systems.
Sensing - Perception
Information Processing
Decision-making
Action Functions
Feedback
Automaticity
Another Basis for Current Models:
Wickens (1992) – Model of HIP influential in Parasuraman et al. (2000) model of types and LOAs.
Perception Decision-making
Response Execution
Working Memory
Long-term Memory
Short-term Sensory Store
FeedbackModel of human information processing (adapted from Wickens (1992)).
Applying Information Processing Models to Automation Design:
– Present ways of classifying functions of human-machine systems.
Note: Entire process is similar to Wickens (1992) approach to systems design.
Mission Analysis
Function Analysis
Task Analysis
Basic System
Concept
Initial Function
Allocation
Dynamic Function
Allocation
Determine “what” to automate?
Determine “how” to automate?
Determine “when” to automate?
Parasuraman et al. (2000) and Endsley & Kaber (1999) models:
Unique Aspects of Approach:
Link LOAs to stages of information processing.
Types of functions considered are general information processing functions.
– Compare with Sheridan & Verplank’s (1978) hierarchy of LOAs – Functions primarily represent action states (“gets”, “starts”, etc.) or choice reactions (“selects”).
Historical function allocation lists technology-centered (Fitts, 1956):
– Advances in understanding of out-of-the (control) loop performance (Endsley & Kiris, 1995; Kaber et al., 1998) shifted focus of lists on human abilities.
– Contemporary taxonomies of LOA (e.g., Endsley & Kaber (1999)) developed by considering performance consequences of automation
Complacency, vigilance decrements, loss of SA, skill decay.
Importance of Research: Link general theories on information processing and
automation to applications.– Parasuraman et al. (2000) - Application of model to air traffic control
(ATC).
Contemporary models of human-automation interaction (HAI) may serve as design rationale for human-centered automation.– Knowledge of performance with systems can be classified according
to model and used to further develop general theory on HAI. Theory may serve to answer several questions:
– What stages of information processing are conducive to automation from human perspective?
– To what extent can stage be automated safely/effectively?– What stages are robust to automation reliability problems?
Theoretical Research Challenges:
Validate processing stages of models in terms of representation of actual functions performed by human-machine systems.
Models can be used for logical function allocation, but practical implications are difficult to predict.– Endsley & Kaber (1999):
Intermediate LOAs moderate workload, maintain situation awareness (SA) and improve performance.
Low level automation produced superior performance, but at cost of SA.
“Good” SA observed under high LOAs.
Hypothesis
Results
Example Model Application:
Applied Endsley & Kaber’s taxonomy of LOAs to advanced commercial aircraft (MD-11) (Tan et al., 2000):– Conducted complex systems
analysis.– Developed high-fidelity
simulation for evaluation of existing system automation.
– Categorized actual modes of automation according to taxonomy.
Flight Control Panel (FCP)Out-of-Cockpit
View
Flight Mode Annunciator (FMA)
Primary Flight
Display (PFD)Navigation Display
(ND)Multi-functionalControl DisplayUnit (MCDU)
Throttle/attitude control
Flight Simulator Displays:
Roles
Level of Automation MONITORING GENERATING SELECTING IMPLEMENTING
(1) MANUAL CONTROL Human Human Human Human
(2) ACTION SUPPORT Human/Computer Human Human Human/Computer
(3) BATCH PROCESSING Human/Computer Human Human Computer
(4) SHARED CONTROL Human/Computer Human/Computer Human Human/Computer
(5) DECISION SUPPORT Human/Computer Human/Computer Human Computer
(6) BLENDED DECISION MAKING Human/Computer Human/Computer Human/Computer Computer
(7) RIGID SYSTEM Human/Computer Computer Human Computer
(8) AUTOMATED DECISION MAKING Human/Computer Human/Computer Computer Computer
(9) SUPERVISORY CONTROL Human/Computer Computer Computer Computer
(10) FULL AUTOMATION Computer Computer Computer Computer
Defining Aircraft Automation in Terms of Taxonomy of LOA:
Taxonomy and example LOA description:
Control Responsibilities Level of Automation
Subsystems Pilot Automation PFD -monitor flight parameters -monitor flight parameters ND -monitor lateral flight plan -monitor lateral flight plan Throttle/ Attitude Control
MCDU -generate/select (explicitly)/ program flight plan manually -edit/re-program (corrective actions) flight plan manually
-implement flight plan selected by pilot -implement flight plan edited by pilot -provide warnings to pilot for corrective actions
FCP
Batch Processing
FMA -monitor modes of operation -display current system status
2Study AA by
considering many points along
continuum of automation.
1
Research to Enhance Basic Understanding of
Automation: Adaptive automation (AA) (or
“when” to automate) primarily studied using laboratory simulations of complex systems.
Majority of AA research investigates concept from binary perspective.– Automation is continuous variable
made discrete for research purposes.
Automation
Full Automation
Manual Control “When?”TnT0
Decision Support
Batch Processing
Decision Support
Supervisory Control
Supervisory Control Manual
Control
Action Support
Binary ApproachContinuous Approach
Study AA in high-fidelity
simulations, or in context.
3
4
5
Considering Models of HAI in Adaptive
Systems Research: Focus of AA research has been on
early stages of information processing:– Do not have understanding of effect of AA
on decision-making and planning.– People may not adapt well to dynamic
allocations of decision functions. Research has not systematically
examined performance and SA effects of dynamic function allocations (DFAs) across stages of information processing.
Study impact of AA on latter
stages of information processing.
Study SA and performance effects of AA
across functions represented in models of HAI.
Study impact of AA applied
exclusively to each stage of information processing.
Parasuraman et al. (1993)
7Consider need to
keep-track of LOAs in AA
studies.
6Compare voluntary,
involuntary and shared DFA management
Final Issues:
– Computer authority shown to reduce excessive cyclings between control modes (Hilburn et al., 1993).
– Automation directed DFAs improve human manual control – operator does not need to evaluate who should be doing what (Kaber & Riley, 1999).
Approach AA research cautiously - complexity of system design may cause mode awareness problems (Sarter & Woods, 1995).
Need to determine who (human or automation) should have authority over DFAs:
Conclusions:
Can’t study AA without considering LOA – concepts “like peas and carrots” (F. Gump).
Taxonomies of LOA provide means for systematically studying complexities of automation.
Need to apply models of HAI to real systems to expand theory to design rationales.– Classify systems and understand underlying factors in
automated system performance.
Need to develop framework of AA research.