1 / 7

Challenge Problems for Multiagent Planning (Competition?)

Challenge Problems for Multiagent Planning (Competition?). Robert P. Goldman. Desirable Features. Opportunity to do projection --- purely reactive approaches not suitable Dead-ends to avoid Prepositioning (umbrella) Optimization of cost and/or time Test robustness under some circumstances

dobry
Télécharger la présentation

Challenge Problems for Multiagent Planning (Competition?)

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Challenge Problems for Multiagent Planning (Competition?) Robert P. Goldman

  2. Desirable Features • Opportunity to do projection --- purely reactive approaches not suitable • Dead-ends to avoid • Prepositioning (umbrella) • Optimization of cost and/or time • Test robustness under some circumstances • Non-free communications • Tunable dynamism (allows us to determine when projection is and is not appropriate) • Heterogeneous agent capabilities

  3. Requirements • Analogy to PDDL: features of domain • Static vs. dynamic environment • Communications through simulator vs. anything goes • Communications that can fail • Agent failure/death • Fully observable state vs. partial observability • Actions • Deterministic • Non-deterministic • Probabilistic • Action durations • Instantaneous • Deterministic • Uncertain

  4. Possible Candidate Problem Family • [Inspired by Coordinators field test] • Map with multiple sites of interest that have • Multiple operations that must be performed at the sites • Operations are linked through preconditions and postconditions • Different agents are needed to perform different operations • Some form of deadline (need not necessarily be metric time) • Enhancements • Partial observability --- need to do a survey before you know what is to be done at a site • (Somewhat) unpredictably get new sites added to the problem • Communications concerns • Push communications through the simulator, instead of making them go direct • Random message loss • Actions that can fail • Agents can get damaged or destroyed (randomly, or in particular situations)

  5. What We Need in a Description • Domain – class of problems • Entity types • State description components • Slots? • Predicates? • Map (if shared) • Action models • Possibly including sensing actions • Method of invoking actions • Requirements list • Objective model • Note: We do not need a plan language • Possibly: common problem components • Specific Problem • Map (if unshared) • Specific objectives (?)

  6. THE END

  7. Title • Bullets • Deeper • And more

More Related