Adapting Memory for Long-Duration Autonomy

Abstract

Our objective is to demonstrate that goal priming and perpetual learning are essential as a system's lifetime and capability increase. Our planned contributions include: (1) a theory of goal priming that combines previous work in goal reasoning and cognitive priming; (2) extending previous work on perpetual learning to long-duration systems; (3) formal and empirical analyses showing how goal priming and perpetual learning combine to support increasingly more capable, long-term autonomy; and (4) demonstrating a system running for 18 continuous months that consolidates experience and retrieves relevant goals based on context in scenarios from two simulated environments: logistics delivery for Foreign Disaster Relief, a Navy-relevant domain, and Minecraft, a challenging domain from Microsoft Research.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Sep 10, 2021
Accession Number
AD1149308

Entities

People

  • Mark Roberts

Organizations

  • United States Naval Research Laboratory

Tags

Communities of Interest

  • Autonomy
  • C4I
  • Space

DTIC Thesaurus Topics

  • Algorithms
  • Artificial Intelligence
  • Autonomous Systems
  • Autonomy
  • Cognition
  • Cognitive Science
  • Databases
  • Environment
  • Humanitarian Assistance
  • Information Systems
  • Learning
  • Machine Learning
  • Open Source Software
  • Platforms
  • Psychology
  • Reasoning
  • Recognition
  • Reinforcement Learning
  • Simulations
  • Simulators

Fields of Study

  • Psychology

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Artificial Intelligence