Agent-Based Computing in Distributed Adversarial Planning

Abstract

This report results from a contract tasking Czech Technical University as follows: This project investigated the problem of adversarial reasoning and planning, i.e. goal-oriented decision making in the presence of other adversarial actors. In contrast to existing approaches, our line of research addresses the problem within the context of complex, asymmetric domains with properties similar to those found in real-world conflict situations (a higher number of parties, asymmetry in party's objectives and resources, huge state space etc.). The focus of the previous project was on fully noncooperative scenarios. The current project took into consideration also (partial) explicit cooperation among individual parties in the scenario. The pursued approach combines theoretical analysis with practical algorithm development with strong emphasis on empirical evaluation using a multiagent adversarial behavior testbed. The project delivered the following specific results: - extended formal framework for adversarial reasoning: The framework refines the concepts defined in our previous work and adds notions related to creating, maintaining and reasoning about coalitions in adversarial and semi-cooperative settings. - extended adversarial behavior testbed: The existing testbed was adjusted to allow negotiations and explicit coordination among the players. New testing scenarios that require coordination of self-interested parties were developed. - agent subset adversarial search: A novel algorithmic scheme for substantial speed-up of generic adversarial-search based algorithms has been developed, implemented and extensively evaluated within the testbed. - sub-game negotiation tree search: The algorithm uses negotiation among self-interested parties to create mutually more beneficial plans. An agent is expected to agree to deviate from its optimal uncoordinated plan only if it - process models for opponent modeling.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Aug 09, 2010
Accession Number
ADA527060

Entities

People

  • Michal Pechoucek

Organizations

  • Czech Technical University in Prague

Tags

Communities of Interest

  • Autonomy
  • C4I
  • Materials and Manufacturing Processes
  • Sensors
  • Weapons Technologies

DTIC Thesaurus Topics

  • Algorithms
  • Artificial Intelligence
  • Commerce
  • Computational Complexity
  • Deception
  • Game Theory
  • Intelligent Agents
  • Knowledge Management
  • Language
  • Machine Learning
  • Military Operations
  • Multiagent Systems
  • Negotiations
  • Probability Distributions
  • Random Variables
  • Sensor Networks
  • Zero-Sum Games

Fields of Study

  • Computer science

Readers

  • Game Theory.
  • Neural Network Machine Learning.
  • Systems Analysis and Design

Technology Areas

  • Space