Autonomous Learning with Probability and Abstraction for Competency Awareness (ALPACA)

Abstract

This report describes research carried out by the Draper team as part of the DARPA Competency-Aware Machine Learning (CAML) program. Draper teamed with subcontractors UT Austin, ASU, and CU Boulder to develop ALPACA (Autonomous Learning with Probability and Abstraction for Competency Awareness), a general framework for competency-aware autonomous agents, particularly those based on reinforcement learning (RL). ALPACA provides insight into RL agent competencies and empowers users to examine and constrain agent behavior, facilitating trust building with human teammates and dramatically improving safety for real-world applications.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 30, 2023
Accession Number
AD1192115

Entities

People

  • Rebecca L. Russell

Organizations

  • Charles Stark Draper Laboratory

Tags

Communities of Interest

  • Autonomy

DTIC Thesaurus Topics

  • Abstracts
  • Accuracy
  • Artificial Intelligence
  • Autonomous Agents
  • Autonomous Systems
  • Classification
  • Computational Science
  • Contracts
  • Dimensionality Reduction
  • Failure Mode And Effect Analysis
  • Generative Models
  • Human-Machine Interaction
  • Human-Robot Interaction
  • Language
  • Lessons Learned
  • Machine Learning
  • Neural Networks
  • Probabilistic Models
  • Probability
  • Recurrent Neural Networks
  • Reinforcement Learning
  • Reliability

Fields of Study

  • Computer science

Readers

  • Artificial Intelligence
  • Defense Technology Research and Development.
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy