Predictive Models of Procedural Human Supervisory Control Behavior

Abstract

Human supervisory control systems are characterized by the computer-mediated nature of the interactions between one or more operators and a given task. Nuclear power plants, air traffic management and unmanned vehicles operations are examples of such systems. In this context, the role of the operators is typically highly proceduralized due to the time and mission-critical nature of the tasks. Therefore, the ability to continuously monitor operator behavior so as to detect and predict anomalous situations is a critical safeguard for proper system operation. In particular, such models can help support the decision making process of a supervisor of a team of operators by providing alerts when likely anomalous behaviors are detected. By exploiting the operator behavioral patterns which are typically reinforced through standard operating procedures, this thesis proposes a methodology that uses statistical learning techniques in order to detect and predict anomalous operator conditions. More specifically, the proposed methodology relies on hidden Markov models (HMMs) and hidden semi-Markov models (HSMMs) to generate predictive models of unmanned vehicle systems operators. Through the exploration of the resulting HMMs in two distinct single operator scenarios, the methodology presented in this thesis is validated and shown to provide models capable of reliably predicting operator behavior. In addition, the use of HSMMs on the same data scenarios provides the temporal component of the predictions missing from the HMMs. The final step of this work is to examine how the proposed methodology scales to more complex scenarios involving teams of operators. Adopting a holistic team modeling approach, both HMMs and HSMMs are learned based on two team-based data sets. The results show that the HSMMs can provide valuable timing information in the single operator case, whereas HMMs tend to be more robust to increased team complexity.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 2011
Accession Number
ADA559041

Entities

People

  • Yves Boussemart

Organizations

  • Massachusetts Institute of Technology

Tags

Communities of Interest

  • Autonomy
  • C4I
  • Cyber
  • Ground and Sea Platforms
  • Human Systems
  • Space
  • Weapons Technologies

DTIC Thesaurus Topics

  • Automata Theory
  • Bayesian Networks
  • Cognitive Science
  • Cognitive Systems Engineering
  • Computational Science
  • Control Systems
  • Data Mining
  • Human Systems Integration
  • Human-Machine Interaction
  • Information Processing
  • Information Science
  • Information Systems
  • Machine Learning
  • Network Science
  • Psychology
  • Supervised Machine Learning
  • Unmanned Systems

Fields of Study

  • Computer science

Readers

  • Computational Modeling and Simulation
  • Distributed Systems and Data Platform Development
  • Instructional Design and Training Evaluation.

Technology Areas

  • Autonomy
  • Autonomy - Human-Robot Interaction