Automatically Inferring Human Machine Interaction Properties and Predicting and Adapting to their Violation
Abstract
The goal of this work is to enable the formal and automated characterization, predictive-monitoring, and adaptation of the rich space of human-machine interactions that occur when a semi-autonomous system is deployed. The technical approach consists of 1) developing specialized parameterizable patterns that can encode aspects of such rich interactions, including their source state, context, time and temporal dependencies, and their uncertainty, 2) developing techniques that can automatically induce the parameters in those patterns into properties utilizing field traces that contain the states and events that constitute those interactions, 3) developing monitors that can check but most importantly predict violations, and 4) investigating techniques that, triggered by the prediction of a violation, can intelligently adapt the interaction protocol to improve the system’s overall behavior. The anticipated outcome is a family of formally defined parameterizable patterns to characterize the interactions between users and systems working together to achieve a goal, a series of automated techniques to instantiate those patterns into properties, predictive monitors in the form of algorithms and prototype tools that can anticipate property violations, and adaptation heuristics based on the property and its supporting metadata to adapt the interactions to better understand the property violation or avoid it altogether. A cross-cutting outcome will be the assessment of each of these technical developments with semi-autonomous ground and aerial systems. The work has the potential to dramatically enhance our understanding of the protocols governing human-machine interactions in deployed systems and, more importantly, to anticipate and reduce the chances that such interactions negatively affect a system’s overall behavior.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 21, 2022
- Source ID
- FA95502110164XX0
Entities
People
- Sebastian Elbaum
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Virginia