Level of automation effects on performance, situation awareness and workload in a dynamic control task

Various levels of automation (LOA) designating the degree of human operator and computer control were explored within the context of a dynamic control task as a means of improving overall human/machine performance. Automated systems have traditionally been explored as binary function allocations; ei ther the human or the machine is assigned to a given task. More recently, i ntermediary levels of automation have been discussed as a means of maintain ing operator involvement in system performance, leading to improvements in situation awareness and reductions in out-of-the-loop performance problems. A LOA taxonomy applicable to a wide range of psychomotor and cognitive tas ks is presented here. The taxonomy comprises various schemes of generic con trol system function allocations. The functions allocated to a human operat or and/or computer included monitoring displays, generating processing opti ons, selecting an 'optimal' option and implementing that option. The impact of the LOA taxonomy was assessed within a dynamic and complex cognitive co ntrol task by measuring its effect on human/system performance, situation a wareness and workload. Thirty subjects performed simulation trials involvin g various levels of automation. Several automation failures occurred and ou t-of-the-loop performance decrements were assessed. Results suggest that, i n terms of performance, human operators benefit most from automation of the implementation portion of the task, but only under normal operating condit ions; in contrast, removal of the, operator from task implementation is det rimental to performance recovery if the automated system fails. Joint human /system option generation significantly degraded performance in comparison to human or automated option generation alone. Lower operator workload and higher situation a awareness were observed under automation of the decision making portion of the task (i.e. selection of options), although human/sys tem performance was only slightly improved. The implications of these findi ngs for the design of automated systems are discussed.