Implications for operator interactions in an agent supervisory control relationship

Abstract

In order to support a human UAV operator during his mission management task, an artificial cognitive agent (ACU), operated on the basis of high-level tasks, is introduced onboard an unmanned automated aircraft. In analogy to human UAV guidance, the ACU does not interact directly with the aircraft, but rather formulates discrete commands for the automation and continually supervises their execution. In accordance with Sheridan's Human Supervisory Control, we describe this relationship between the agent and the underlying automation by the term of Agent Supervisory Control. In this role, the ACU performs Supervisory Functions, with which the tasks of the operator are supplemented, reviewed and finally decomposed into single actions. These actions are then performed using the onboard conventional automation. The agent considers the automation functions of the UAV as capabilities that are managed and incorporated into its planning and execution process in accordance to their availability and requirements. During task execution, the ACU supplies the operator with information, as to keep him in charge of the aircraft at all times. The extent, form and frequency of the information depend on the necessities of the specific situation and the current task. The information requirements of the operator and a possible operator information-overload due to its detail must be weighed against each other.

Experimental evaluation of varying feedback of a cognitive agent system for UAV mission management

“…Nowadays, military Remotely Piloted Aircraft System mission management is in focus of research (Clauss & Schulte, 2014;Theissing & Schulte, 2015). In modern UAV-systems, conventional automation (i.e., auto-flight systems) relieves the operator of high-frequent sensor-motor tasks and improves precision and performance for mission execution.…”

“…Tasks might be military reconnaissance missions. TBG stems from an inter-human delegation relationship and relieves the supervisor from the tedious task to derive automation action instructions from intentions (Clauss & Schulte, 2014).…”

“…In an experimental campaign the impact of the agent's feedback behavior on planning, commanding and re-planning of the operator is examined. The developed cognitive agent (Clauss & Schulte, 2014) with advanced feedback abilities is evaluated with respect to human performance, re-plan capability and human trust in the agent. In our experiment we compare the full feedback system to a configuration with reduced feedback information.…”