Human Interaction with Control Software Supporting Adjustable Autonomy

Reductions in operations personnel and the availability of more capable automation technology have resulted in increased automation of space control systems. One of the challenges of deploying autonomous systems is ensuring that people can interact effectively with them. An important strategy for such interaction is providing for different levels of control autonomy, or adjustable autonomy. At TRACLabs/Metrica, we are developing software for human interaction with control software supporting adjustable levels of autonomy. Our domain of application is the control of space systems and robots for NASA's manned space exploration program at the Johnson Space Center. Traditionally test and flight operations have been manually intensive. Flight controllers monitor astronauts directly controlling the operation of space vehicle systems, robotics, and life support systems. The use of automated software changes the role of humans in command and control. Low level data monitoring and control is replaced by the supervision of autonomous control and by joint human-computer task performance. These new human tasks require new types of information and new ways of presenting this information, as well as new ways of intervening in autonomous operations. Additionally, these changes in operations have increased the likelihood of humans supervising multiple automated systems that are physically dispersed and functionally diverse. It is important to provide an integrated view of system operations and to support access to such information from remote locations. Intelligent information processing is needed to derive and integrate heterogeneous control data in support of these new human tasks. And an architecture is needed that supports remote human interaction with distributed, autonomous control systems. We use the Three Tier (3T) distributed control architecture [3] to automate system control. The 3T architecture consists of three parallel tiers of control processing: • Planner: determines tasks required to meet control objectives and monitors task execution • Sequencer: reactively activates sequences of situated skills to accomplish planned tasks • Skill Manager: manages skills interfaced to instrumentation to achieve desired world states We are developing a fourth layer of software the interaction layer for human interaction with control software supporting adjustable autonomy. The interaction layer collects, processes, and integrates information from each tier of the agent architecture to provide a human supervisor with a view of control activities integrated with the intended effects of control. It also provides the human supervisor with the ability to interact with each tier of the architecture to perform such tasks as mixed-initiative planning, situation assessment, anomaly response, and manual override of automation. To effectively use automated control software, the human must be able to monitor autonomous activities easily and, with little effort, be able to intervene if the situation calls for it. We have developed user interface designs that support minimal human involvement during routine operations, and that assist humans in actively participating in control during novel or complex operations. We have designed user interfaces that support a human supervisor in (1) understanding automated control actions and their environmental consequences, (2) accessing the archived details of these control activities easily, and (3) intervening in control as needed. We have made progress in adding the interaction layer to the 3T architecture. The interaction layer features distributed processing with client-server process communication. It includes common data models that characterize both the information that must be exchanged between the control system and the human and the representation of that information best suited for supervisory tasks. We have built reusable user interface software that interfaces with the sequencer and skill manager tiers of 3T. The software developed for the sequencer user interface provides a suite of applicationindependent display and logging tools. Each tool can be From: AAAI Technical Report SS-99-06. Compilation copyright © 1999, AAAI (www.aaai.org). All rights reserved.