David B. Kaber Mississippi State University
Emrah Omal Lucent Technologies
Mica R. Endsley SA Technologies
Recent literature ( Endsley and Kaber (in press); Endsley and Kiris, 1995; Milgram et al., 1995) on advances in automation for complex, dynamic systems in general, and telerobots (remote controlled robotic manipulators) specifically, has presented taxonomies of levels of automation (LOAs) detailing control function allocation schemes that may improve systems performance over that resulting from traditional automation. (Traditional automation is considered to be the implementation of technology based on its capabilities, but lacking in consideration of the effects of its application on a human operator.) The taxonomies have also been put forth as vehicles by which to address out-of-the-loop performance and safety problems along with their underlying causes. Root causes to performance problems include operator vigilance decrements and complacency, which may both contribute to problems with operator situation awareness (SA). Most of the taxonomies offer intermediary LOAs falling somewhere between manual control and full automation. These levels are intended to maintain both human and computer involvement in active systems control for improving operator SA (critical to failure recovery and system safety) and increasing system performance (through computer data processing). The taxonomies present levels by identifying or describing the roles that the human operator and computer are to play in controlling a system.
Endsley and Kiris ( 1995) presented a taxonomy of LOA developed in the context of the use of expert systems to supplement human decision-making for automated systems control. They identified five functions that either a human operator or expert system could play including suggest, concur, veto, decide, and act. They offered five LOAs by structuring allocation of these roles to both servers ranging from "Manually" to "Full Automation." Endsley and Kiris ( 1995) empirically assessed the effect of the LOAs in this taxonomy on out-of-the-loop performance problems and SA in a simulated automobile navigation task, revealing it to be more significant under fully automated conditions than under intermediate LOAs. They found that using lower LOAs, which maintain human operator involvement in active control, was beneficial to SA and subjects were better able to perform tasks manually when needed.
Endsley and Kaber (in press) present a taxonomy of LOAs developed by allocating to a human and/or computer generic control system functions including monitoring, generating, selecting and implementing based on the capabilities of each server to perform the functions. Endsley and Kaber (in press) formulated 10 LOAs feasible for use in the context of teleoperation. The levels have been assessed as to their effect on human-machine system performance, and operator SA and workload in a dynamic control task. They have also been studied as to their potential for facilitating a smooth transition (in terms of performance) between normal operations and simulated automation failures. Endsley and Kaber (in press) found human- machine system performance to be enhanced by automation that provided computer aiding or control in the implementation aspect of the task. With respect to performance during failure modes, the authors found human control to be significantly superior when preceded by functioning at LOAs involving the operator in