TITLE: human-automation interaction AUTHOR: QUENCY DATE: 10/11/2013 12:23:41 AM CATEGORY: 人因工程設計 STATUS: publish ---- BODY:
Human-Automation Interaction Research: Past, Present, and Future
(April 2013 1 ergonomics in design)
A range of perspectives on human-automation interaction may serve to sustain future performance in the area.
By Peter A. Hancock, Richard J. Jagacinski, Raja Parasuraman, Christopher D. Wckens, Glenn F. Wilson, & David B. Kaber
Science is redolent with predictions of the future grounded in rationales of the past, and some of these predictrons have been both prescient and informative (Bartlett, 1962; for a commentary, see Hancock, 2008). Those who have lived through the development and evolution of particular ideas are often well equipped to speculate about future impacts. The main purpose of this work was to capture
the knowledge and opinions from a group of researchers who have been extensively involved in the field of human-automation interaction and especially the emergent form of adaptive automation, which replaced the one-time, static allocation procedures characterizing a major part of human factors/ergonomics (HF/E) science in the immediate post—World War II interval (e.g., Pitts, 1951).
ln several ways, the present work builds on previous observations about complex system function allocation and human perlorinance implications presented at an earlier annual meeting of the Human Factors and Ergo-nomics Society (see Sheridan, Hancock, Pew, Van Cott, 8ttW, 1998, for a report). The question at that time was whether funetion allocation could be prescribed in a rational manner, with several opinions for and against. What has occurred in HF/E science since that time is a resounding respottsc of “for” from researchers and the development of approaches to adaptive automation; that is, rational assignment of system functions to human and machine on a real-time basis for worldoad management and performance optiinization.
What follows are the differing perspectives of each researcher involved in the development of seminal theories and observations on humam-automation interaction, with a focus on adaptive forms. We begin with an evaluation by Jagacinski of the utility of theoretical models (in control) for describing human performance with complex automated systems involving cognitive tasks.
This section is followed by an examination of human behavior implications of levels of automation by Wickens and identification of findings counter to conventional wisdom as well as approaches that may
serve to sustain performance in varying automation conditions. The focus then shifts to the origins of adaptive automation and recounting of key developments toward human-centered automation, as covered by
Hancock and Parasuraman.
Finally, Wilson provides coverage of methods and issues in operator functional state classification for adaptive system control. The article culminates with a look forward and identification of some opportunities and challenges in the area of human-automation interaction, as addressed by Hancock and Kaber.
COMPARISON OF MODELS OF DECISION AND CONTROL TASKS
By Richard Jagacinski
Important parallels can be drawn between behavioral models of decision tasks and control tasks involving complex automated systems. A control task can be considered as a sequence of decisions of how to influence a dynamic system or process. The effects of each decision or action determine the context for the next decision in the sequence. There are strong parallels between regression models of decision making and classical control models of tracking in aircraft-piloting and car-driving tasks (e.g., McRuer & Jex, 1967;McRuer,Allen Weir, & Klein, 1977), between Bayesian decision models and models of the optimal esatimation of noisy signals for vehicular control (e.g., Kleinman, Baron, & Levison, 1971), and between heuristic models of decision making and state space models of discrete control, such as the steering of large ships (e.g., Veldhuyzen & Stassen,1977).
Analogous to signal detection theory (e.g., Swets, 1982), control theory has permitted researchers to separate strategicJagacinski & Flach, 2003). However, unlike most common uses of signal detection theory, control theory does not typically categorize single actions or decisions in a sequence as correct or incorrect but, instead, describes the stability and efficiency of the extended sequence for reaching desirable goal states over time.
For example, how aircraft control dynamics influence the potential for oscillations or divergences from intended flight trajectories hasheen extensively studied. The redesign of aircraft that exhibit such tendencies may involve lowering control sensitivities, reducing lags, and/or filtering structurally resonant frequencies from the pilot's control movements (Committee on the Effects of Aircraft-Pilot Coupling on Flight Safety, 1997).
Related to the comments of Woods (Sheridan et al., 1998) on the historical ftinction allocafion problem, policy decisions in sociotechnological systems can be extremely difficult because of the complexity of system dynamics (e.g., Forrester, 1970). Such dynamics can also change with the introduction of new technology and lead to system instability.
Instances of people's collective inability to reach consensus on recognizing and/or controlling vehicular, financial, and ecological instabilities before they become harmful, such as the 2008 crash of the U.S. housing market, suggest that the quantitative predictions and qualitative intuitions based on control theory need tube more effectively utilized to debate options and to inform system designers on policy decisions involving humans and technology. The study of manual control (e.g., Jagacinski & Flach, 2003) can provide a relatively friendly introduction to relevant dynamic concepts for addressing such problems.
TRADE-OFFS IN HUMAN-AUTOMATION INTERACTION: ARE THEY INEVITABLE?
By Christopher D. Wickens
A long-held conventional wisdom is that a greater degree of automation in human-in-the-loop systems produces both costs and benefits to performance. Here degree of auto-mation is defined both by higher levels of system autonomy (more machine authority on the Sheridan and Verplank, 1978, scale) and application to later stages of information processing (more automated action selection and execution authority on the Parasuraman, Sheridan, & Wickens, 2000, taxonomy).
The major benefit of automation is performance in routine circumstances. Costs result when automation (or the systems or sensors controlled by automation) "fail" and the human must intervene, as in the crash of Air France Flight 447 (see Wise, 2011), This degraded failure response is assumed to result jointly from complacency (operator overreliance on automation) and from elimination of the called generation effect; that is, when operators are not involved in generating action alternatives. Together, these two effects define the loss of situation awareness (SA) at Level 1 (complacency) and Level 2 (the generation effect: one has reduced memory for actions one did not generate). based on Endsley's (1995) model.
Along with SA change, a second inferred variable is the reduction of workload that comes with greater degrees of automation, as humans are left with less to do. Thus the changes in these four variables (routine and failure performance, SA, and workload), produced by changes in degree of automation imply an elegant, but perhaps overly simplified, model. This conceptual model has historic threads that reach back through Billings's (1991) elegant work on human-centered automation, through early empirical and theoretical work on accidents induced by flight deck automation (Wiener & Curry, 1980) and the broader realm of other industrial environments (Rasmussen & Rouse, 1981).
offs are far from inevitable. In 2010, Wickens, Li, Santamaria, Sebok and Sorter (2010) reported that increased degree of automation? Discovering these features provides the Key to establishing useful guidelines for human-automation interaction. some hints suggest that effective intuitive displays can buffer high levels of automation from human performance costs when things fail, perhaps capitalizing on principles of ecological interface design. For example. Label, Perry, Segall, McClernon, and Prinzel (2006) demonstrated that auditory and visual cuing of adaptive automation with supervisory control of a telerover served to improve operator performance and reduce decrements in SA attributable to out-of-the-loop unfamiliarity. A second source of mitigation may be effective training: training to "expect the unexpected" as well as training in understanding of automation logic.
THE FOUNDATIONS OF ADAPTIVE AUTOMATION IN PHYSIOLOGICAL THEORIES
By Peter A. Hancock
The idea of having machines adapt to the cognitive and physical demands of users in a momentary and dynamic manner (i.e., adaptive automation) is one of the more important ideas its the history of humam factors/ergonomics. It has served to propel the field forward from its static incarnation of function allocations, seals as was represented its the now famous Fitts list (Fitts, 1951), into something that has now exploited the extensive advantages of the computational revolution (Hancock & Scallen, 1996). However, the notion of dynamic adaptation itself was not really something new.
Indeed, the concept of adaptation was one that has permeated the annals of biology, even before its most famous formalization by Charles Darwin. But its biology, adaptation is accomplished almost exclusively by changes its the organism itself. Typically, these changes occur in proportion to the lifetime of the individual, and, thus, gradually, each species itself changes. But in the natural world, change happens on many time scales.
Can organisms change more directly its response to purposive elements in the environment? Although it is true that many animals use "tools" in the crudest sense of the word, the notion of orthotics being purposively produced by a species, which then itself subsequently co-adapts, is confined solely to human beings. Indeed, I would argue that this is one, if not the, central hallmark of humanity.
Formally, this co-adaptation occurs at a frequency that is derived from an integration of the respective time scales of change, as represented by variation rate its the organism and the tool, respectively. This characteristic interchange between humans and the machines they create, which I have elsewhere labeled the self-symbiotic species, is what makes human beings unique. Adaptive automation has served to take this, our form of hybrid development, one step further into the future.
Adaptation in human-machine systems was, from my perspective, first broached systematically at a behavioral level by Rouse, who sought to turn conflicting intelligences into cooperating ones (see Rouse, 1988). His work on the frustrations of dynamic incarnations of the Fitts list and MABA-MABA (men are better at-machines are better at)-like specifications began this important revolution, which is still contro-versial and playing out (see Dekker & Woods, 2002).
based insights meshed with my background in human biology as a basis for positing task load redistribution through analysis of observed physiological and neurophysiological signals. Our first published work on this approach to adaptive automation emerged shortly aftenvard. For my own part, the publication was inspired by the seminal insights of scientists such as Bernard, Cannon, and Selyc on stress-related regulation of central and peripheral nervous system operations (sec Hancock, Chignell, & Loewenthal, 1985). Others focusing on practical implementation have since advanced our original theoretical observations on such methods. These efforts led to and resulted in major research programs, including augmented cognition (Schmorrow, 2005) and neuroergonomics (Parasuraman & Rizzo, 2007), which are founded on these principles (see Parasuraman Hancock, 2004).
Today, the partnership between humans and machines in complex systems is becoming ever more intimate. The significant advances in recording and interpreting physiological signals, especially those from the active central nervous system, have rendered what was once only a theoretical concept into a practical field technology. Not only can tasks be parsed according to which partner might be best able to deal with demands on a momentary basis, but one can also use technologies to provide to operators much wider access to computer-based information. Examples of such adaptive automation technologies include the rotary pilots associate (RPA; see Miller and Hansen, 1999, for an assessment), which provided Apache pilots with the capability to negotiate with automation on flight task allocations and intelligently identified critical operation information to support performance. Thus, adaptation has now extended beyond the sharing phase to embrace a wider vista of augmentation, but in a true sense, this is adaptation at the "next level."
In the ongoing marriage of mind and machine, the neurophysiological portal through which a fuller integration could occur was opened marginally two decades ago by some promising theoretical conceptions. The present generation is passing rapidly through that conceptual window. However, this race for a greater human-machine intimacy may be more than simply another step along the unique road of history. For if the present vector of self-destructive progress continues, it may be that this avenue of development is the one that holds the greatest (some would say, only) promise for salvation.
TRACING THE HISTORY OF ADAPTIVE AUTOMATION
By Raja Parasuraman
Related to Hancock's account, the view that automation should be designed and implemented in an adaptive manner can be traced back to Licklider (1960), an early pioneer in computer science, following on the heels of Fitts's (1951) discussions of function allocation, as part of human factors science. In that era, automation was mainly applied to physical functions. As the trend toward greater computerization of the workplace continued, however, the possibility arose for automation of decision making and other cognitive functions.
Sheridan set the stage for the next development with his model of supervisory control. in which a human operator controlled a physical process through an intermediary com-puter (Sheridan & Verplank. 1978). A decade later, researchers such as liancock et al. (1985), Parasuraman (1987), and Rouse (1988) suggested that such systems could benefit human pertbrmance if the nature of the interaction between the human and the computer was adapted to task or contextual demands.
Although these early conceptual and analytical efforts were important, what was missing was a body of empirical research that could test and validate the adaptive automation concept. Fortunately, funding for such research was initiated by the U.S. Navy, in the form of the Adaptive Function Allocation for Intelligent Cockpits (AFAIC) program in 1980, the basis for which was described in a technical report by Parasuraman, Bahri, Deaton, Morrison, and Barnes (1992). A key player in the development of that program was Michael Barnes, who has continued his interest in the adaptive automation concept to this day in his work on human interaction with unmanned vehicles (Barnes, Parasuraman, & Cosenzo, 2006). The AFAIC program led to a period of empirical research on the performance benefits and potential costs of adaptive automation (for reviews, see Inagaki, 2003; Scerbo, 2001).
Empirical research on adaptive automation has demonstrated its benefits with respect to mitigating, at least partly. Some of the costs associated with human-automation interaction. such its unbalanced mental workload) (Hilburn, Iorna, Byrne, & Parasuraman, 1997), complacency (Parasuraman, Moulotta. & Molloy, 1996), and reduced SA (Kaber, Wright, 8; Sheik-Nainar, 2006). At the same time, many issues concerning adaptive autotnation remain to be resolved, including the evaluation of different methods for adaptation, the relative merits of system-versus user-based adaptation, and refinement of neuroergonomic measures and algorithms. Recent neuroergonomics research suggests that physiological measures--such as transcranial Doppler sonography for cerebral blood flow velocity (Shaw, Parasuraman, Guagliardo, & de Visser, 2010)--electroencephalography (Christensen, Estepp, Wilson, & Russell, 2012; Wilson & Russell, 2007), and heart rate (Ting et al., 2010), may be sensitive to unpredictable task load changes and useful for the application of adaptive automation in command and control tasks requiring information acquisition, analysis, and action implementation.
ONE THE USE OF PHYSIOLOGICAL MEASURES TO DETERMINE OPERATOR FUNCTIONAL STATE IN IMPLEMENTING ADAPTIVE AIDING
By Glenn F. Wilson
Adaptive aiding has been found to improve system performance by providing automation when an operator needs it. Because the aiding should be provided only when required to help a compromised operator, it is critical that the functional state of the operator be continuously monitored, otherwise the "aiding" may not be provided when needed or could be provided at the wrong time, interfering with the operator and resulting in increased demands on his or her cognitive resources.
As Parasuraman mentioned, there is a need for additional studies on methods of implementing adaptive automation. Determining when to provide adaptive aiding is, especially difficult in highly automated systems in which the operator's role is primarily one of monitoring the system, as in advanced automated aircraft cockpits. In these situations, there is very little oven operator performance to observe in determining whether the operator could benefit from aiding. One method of determining when to provide aiding is to monitor the functional state of operators using data front their physiology. Previous research has shown that these physiological sig-nals can provide accurate estimates of operator functional state (OFS; Wilson & Russell, 2003a). This finding has been reported with regard to conditions such as mental workload. Brain, cardiac, and eye signals are particularly relevant features when deriving accurate estimates of OFS (Berka et al., 2004; Wilson & Russell, 2003b).
To accurately determine the functional state of the operator, several physiological measures are combined via classifier. Various classifiers have been used to derive estimates of OFS, such as artificial neural networks, discriminant analysis, and support vector machines. The classifier is typically trained using data that represent the cognitive states of interest from each operator separately, or it is trained on previously obtained data from a group of similar operators. The trained classifier is then continuously provided with physiological data during task performance, which it uses to make estimates of the functional state of the operator. When the classifier determines that the operator requires assistance, the system is notified so that it can provide the appropriate automation. As Miller and Karmen (1999) said of the RPA, the system provided pilots with the necessary information for effective flight control at critical times. This type of system requires embedding kowledge of the mission scenario (level of job demands) in the adaptation technology so that suitable aiding can be provided at any time during the mission. Using these procedures, researchers have reported significant improvements in complex task performance (Wilson & Russell, 2007) for various applications, including unmanned (highly automated) aerial vehicle control. These results suggest that physiologically determined adaptive aiding may he included in future systems. Improvements in the areas of sensors, classifier, and computing speed support this development.
SUMMARY AND CONCLUSIONS
Theories and principles of human-automation interaction and adaptive automation have come a long way since their origins in the 1950s and early '60s. From concerns about the real-world application of static function allocation (Fuld, 1993) emerged the promise of a solution through adaptive allocation. Barriers to the practical realization of such structures came through definitions of types and levels of human-centered automation along with empirical research on conceptual forms of adaptive automation.
Demonstrations of model-based adaptive automation and benefits for human performance, workload. SA, and so on motivated work on more precise methods of adaptation to operator cognitive and physial needs. Physiologically based approaches to CMS assessment and connections with methods of task redistribution set the stage for near-real-time operator aiding under workload. Methods of neurophysiological imaging and neurophysiological techniques were a focus of the Augmented Cognition Program (Schmorrow, 2005), and research results provided a boost for adaptive automation applications. The emergence of neuroergonomics (Parasuraman & Rizzo, 2007) as a major area of research and practice in human factors/erganomics has also continued this trend.
Now we are moving from progressively greater individual and task-related diagnosficity for adaptive function allocation toward a fuller agenda of human-machine intimacy. The benefit of our history is an understanding of previously unpredicted relationships of cognition and automation that appear to be mediated by human-machine system interface design and specialized approaches to training. Indeed, with certain physical orthotics and prosthetics note resident in the human individual, as well as emerging brain-computer interfaces (BCI; energized by the medical community), we are beginning to move from the birth of such conceptions into their viable childhood. For example, BCls have been developed for adaptive robotic system control whereby the system responds to the "thoughts" of the human (Gergondet et al., 2011), which are mediated by workload states.
However, there remains a need to apply established theories (including control) to quantitatively predict optimal states of such human-machine symbiosis and provide a basis for ensuring complex dynamic system stability and performance in a range of operating conditions. Human factors engineers can provide designers with guidance on how to develop effective and intuitive forms of such interfaces for preventing automation costs, as Wickens suggests, and how to develop effective training programs for addressing unknown operating circumstances. What the fully mature hybrid human mind and automated machine integration will actually look like and how it will function is the reality the next generation will see and live in.
REFERENCES
Barnes, M., Parasuraman, R., tk Cosenzo. K. (2006). AdaPtwe automat.. for military robotic systems. In Uninhabited military vehicles: Minton factors issues in augmenting the force (NATO Pub. No. RTOTR-HFNI-0711, pp. 420-4.). Brussels, Belgium: NATO.
Bartlett, F. C. (1962). The future of ergonomics. Ergonomic, 5, 505-511.
Berko, C.. Levendowski, D. J., Cvetinovic, M. M., Petrovic, M. M., Davis, G., Lumicao, M. N., ... Olmstead, R. (2004). Real-time analysis of EEG indexes of alertness, cognition, and memory acquired tsith a wireless EEG headset. International Journal of Human-Computer Inferocrion. 17. 151-170.
----- PING: TITLE: http://michiganladder.com/shoes/how-to-lace-nike-air-max-2014-CPFl.html URL: http://michiganladder.com/shoes/how-to-lace-nike-air-max-2014-CPFl.html DATE: 06/06/2015 02:49:12 AM http://elkatamada.ru/js/eastbay-review.html... ----- PING: TITLE: http://arfenya.com/js/oakley-half-jacket-xlj-review.html URL: http://arfenya.com/js/oakley-half-jacket-xlj-review.html DATE: 06/05/2015 05:39:28 PM http://www.happy-look.ru/js/oakley-spare-lenses.html... ----- --------
請先 登入 以發表留言。