Continuous Learning and Explanation for Goal Reasoning

Abstract

Unmanned autonomous military vehicles operate in complex, open, and often adversarial environments in which unexpected situations (impasses or opportunities) can arise during mission execution. It is impossible to pre-encode all knowledge concerning how an agent that controls such a vehicle should respond in all possible situations. We propose to automatically learn deliberative control knowledge for such an agent. To adapt competently in challenging environments, an agent should monitor its environment and,when needed, modify the goals/objectives it pursues (i.e., in addition to its plans, schedules or actions for achieving these goals). To address this, we have developed and tested several goal reasoning (GR) agents with these capabilities. Our GR process model includes functions for (1) Situation Assessment and (2) Decision Making components that operate on sets of long-term (e.g., actions that the vehicle can perform) and short-term data models (e.g., current belief state. Decision Making involves selecting and applying functions (called strategies) to selected goal.Prior work on goal selection had two limitations. First, it relied on hand-coding, which limited an agent’s adaptability. A more robust approach is to learn the goal selection logic from observations, available models, and experience. Earlier goal reasoning agents that used machine learning techniques did not generate (external) explanations of learned knowledge to users. Rather, these techniques (e.g., to learn environment models or relative goal utilities) generated (internal) explanations in response to unexpected situations (i.e., due to observations or actions), which were used by the goal reasoning agent to hypothesize their cause.

Document Details

Document Type
DoD Grant Award
Publication Date
May 30, 2018
Source ID
FA23861814005

Entities

People

  • Claude Sammut

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of New South Wales

Tags

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Team-Based Human-Centered Cognitive Task Decision Making and Information Performance.
  • Theoretical Analysis.

Technology Areas

  • AI & ML
  • AI & ML - Autonomous Systems
  • Autonomy