Multi-Agent Systems Lecture IV
• Dr. Nestor Velasco Bermeo,
• Researcher CONSUS (Crop OptimisatioN through
Sensing, Understanding & viSualisation), • School of Computer Science
• University College Dublin (UCD)
Lecture IV Learning Objectives
❑ Review the difference between reactive and deliberative agent architectures
❑To understand the Belief-Desire-Intention Architecture
❑To understand the different classes of Agent Communication. ❑To understand the different classes of Commitment Strategies. ❑To understand the principles and importance of Speech Acts
Why Deliberative Architectures
•Agents with reactive architectures:
•Can’t reason over hypothetical elements or situations.
•Perform poorly in environments where actions can’t be ignored if proven to be unwise.
•Can’t organize activities over time to coordinate with other agents.
• Represents simple behaviour.
•It’s complicated to present an “intelligent” behaviour from a purely reactive architecture.
Simple Reactive Structure
Another Example… (MES)
Mars Explorer System
• Implements Brooks’ Subsumption Architecture • Hierarchy of task accomplishing behaviours
• Follows simple-rule structure
• Competing for control
• represented as augmented finite state machines (AFSM)
•Triggered when an input surpasses a threshold
• lower level modules can inhibit those in higher levels • modules are grouped and placed into layers
Deliberative Architectures
•BDI – Belief Desire Intention
•PRS- Procedural Reasoning Systems
•IRMA – Intelligent Resource-Bounded Machine Architecture
Belief Desire Intention Architecture
• Employed in the development of Reflective Systems.
• Based on Michael Bratman’s philosophical model of human practical reasoning.
• The term BDI is attributed to Rao and Georgeff (1992).
• Models the reflective process in terms of the interplay between these three mental attitudes.
• Implemented model of practical reasoning agents
Procedural Reasoning Systems (PRS)
•Each agent is equipped with a plan library
Beliefs
agent
Plan Library
Action
• Such
procedural knowledge.
•No plans == No Options
• Agents with PRS posses BDIs
agent’s
Sensor
library
represents
Interpreter
explicit
Desires
Intentions
Intelligent Resource-Bounded Machine Architecture (IRMA)
•Based on the following data structures: • Plan library
• Beliefs
• Desires
• Intentions
• But also:
• Reasoner
• means-end analyser
• Opportunity analyser (environment monitor & Option generator) • Filtering process (compatibility)
Intentions (structured into plans)
surviving options
intentions
Plan Library
Means-end reasoner
Beliefs
Reasoner
Action
Opportunity analyser
Filtering Process
Compatibility Filter
Filter Override Mechanism
Perception
Deliberation process
Desires