The effects of level of automation and adaptive automation on human performance, situation awareness and workload in a dynamic control task

Abstract

This paper extends previous research on two approaches to human-centred automation: (1) intermediate levels of automation (LOAs) for maintaining operator involvement in complex systems control and facilitating situation awareness; and (2) adaptive automation (AA) for managing operator workload through dynamic control allocations between the human and machine over time. Some empirical research has been conducted to examine LOA and AA independently, with the objective of detailing a theory of human-centred automation. Unfortunately, no previous work has studied the interaction of these two approaches, nor has any research attempted to systematically determine which LOAs should be used in adaptive systems and how certain types of dynamic function allocations should be scheduled over time. The present research briefly reviews the theory of humancentred automation and LOA and AA approaches. Building on this background, an initial study was presented that attempts to address the conjuncture of these two approaches to human-centred automation. An experiment was conducted in which a dual-task scenario was used to assess the performance, SA and workload effects of low, intermediate and high LOAs, which were dynamically allocated (as part of an AA strategy) during manual system control for various cycle times comprising 20, 40 and 60% of task time. The LOA and automation allocation cycle time (AACT) combinations were compared to completely manual control and fully automated control of a dynamic control task performed in conjunction with an embedded secondary monitoring task. Results revealed LOA to be the driving factor in determining primary task performance and SA. Low-level automation produced superior performance and intermediate LOAs facilitated higher SA, but this was not associated with improved performance or reduced workload. The AACT was the driving factor in perceptions of primary task workload and secondary task performance. When a greater percentage of primary task time was automated, operator perceptual resources were freed-up and monitoring performance on the secondary task improved. Longer automation cycle times than have previously been studied may have benefits for overall human-machine system performance. The combined effect of LOA and AA on all measures did not appear to be 'additive' in nature. That is, the LOA producing the best performance (low level automation) did not do so at the AACT, which produced superior performance (maximum cycle time). In general, the results are supportive of intermediate LOAs and AA as approaches to human-centred automation, but each appears to provide different benefits to human-machine system performance. This work provides additional information for a developing theory of human-centred automation.

Drivers’ Reliance on Lane-Keeping Assistance Systems as a Function of the Level of Assistance

“…Kaber and Endsley (2004) found that improvements in SA with intermediate levels of automation in a dynamic control task were not related to reductions in mental workload. Endsley and Kiris (1995) also found that changes in level 2 SA (situation comprehension) with varying levels of automation in a decision task were unrelated to participants' reported mental workload.…”

Impact of automation: Measurement of performance, workload and behaviour in a complex control environment

“…Durso & Sethumadhavan, 2008;Endsley & Kiris, 1995;Kaber, Onal, & Endsley, 2000;Kaber, Perry, Segall, McClernon, & Prinzel, 2006) and workload (e.g. Kaber & Endsley, 2004;Kaber et al, 2006;Kantowitz, 1994). The levels of automation identified in the models can be used to distinguish levels of independent variables in experimental designs; if sufficient levels are defined, the effect of automation can being to be described on a continuum.…”

“…This may be due to the need to assimilate greater quantities of information (Macdonald, 1999) or because monitoring of automation becomes burdensome (Kaber & Endsley, 2004;Warm, Dember, & Hancock, 1996). It is also the case that it is often easier to automate information acquisition and action implementation, leaving the cognitive load unchanged for operators.…”

“…This may be due to the need for operators to monitor the automation during information analysis and decision making phases and continue to make their own decisions as a basis for comparison with the automation. Kaber and Endsley (2004) also suggested that the very act of monitoring an automated system may increase workload, perhaps due to the effort of remaining vigilant (Warm, Dember, & Hancock, 1996). Relatively few studies have directly evaluated workload and levels of automation during disruption, but Kantowitz (1994) found that automation may increase workload during incidents.…”

“…Numerous lab-based studies in the field of human factors have been undertaken to investigate the effects of automation and these have often found the benefits to be less clear-cut than might be expected (Parasuraman & Riley, 1997). For example, situation awareness may be reduced under high levels of automation (Kaber & Endsley, 2004) and workload may be increased under abnormal circumstances (Kantowitz, 1994). The level of reliability of automation is crucial, with a level below 70% believed to be worse than no automation (Wickens & Dixon, 2007).…”