Skip to content

UsingFIPAACL

RoiArthurB edited this page Feb 22, 2024 · 1 revision

Using FIPA ACL

GAMA allows modelers to provide agents the capability to communicate with other agents using FIPA Communication Acts (such as inform, request, call for proposal...) and Interaction Protocols (such Contract Net Interaction Protocol, Request Interaction Protocol).

To add these capabilities to the chosen species, the modeler needs to attach the fipa skill: it adds to agents of the species some additional attributes (e.g. the list of messages received) and available actions (e.g. the possibility to send messages given the chosen Communication Act).

The exhaustive list of available Communication Acts and Interaction Protocols is available from the technical description of the fipa skill page. Examples can be found in the model library bundled with GAMA (Plugin models / FIPA Skill).

Table of Contents

Main steps to create a conversation using FIPA Communication Acts and Interaction Protocols

  1. Attach the skill fipa to the agents' species that need to use Communication Acts
  2. An initiator agent starts a conversation with some agents: it chooses the Interaction Protocol and starts it by sending the first Communication Acts of the protocol
  3. Each agent involved in the conversation needs to check its received messages and respond to them by choosing the appropriate Communication Act.

Attach the fipa skill to a species

To attach the fipa skill to a species, the modeler has to add it in the skills facet of the species statement (in a way similar to any other skill).

species any_species skills: [fipa] {
   ...
}

Agents of any species can communicate in the same conversation. The only constraint is that they need to have the capabilities to receive and send messages, i.e. to have the skill fipa.

Species can have several attached skills: a single species can be provided with both the moving and fipa skills (and any other ones).

This skill adds to every agent of the species:

  • some additional attributes:
    • conversations is the list of the agent's current conversations,
    • mailbox is the list of messages of all types of performatives,
    • requests, informs, proposes... are respectively the list of the 'request', 'inform', 'propose' performative messages.
  • some additional actions, such as:
    • inform, accept_proposal... that replies a message with an 'inform' (respectively 'accept_proposal' performative message).
    • start_conversation that starts a conversation with a chosen interaction protocol.
    • end_conversation that replies a message with an 'end_conversation' performative message. This message marks the end of a conversation. In a 'no-protocol' conversation, it is the responsibility of the modeler to explicitly send this message to mark the end of a conversation/interaction protocol.
    • reply that replies a message. This action should be only used to reply a message in a 'no-protocol' conversation and with a 'user-defined performative'. For performatives supported by GAMA, please use the 'action' with the same name as the 'performative'. For example, to reply a message with a 'request' performative message, the modeler should use the 'request' action.

Initiate a conversation

An interaction using an Interaction Protocol starts with the creation of a conversation by an agent, using the start_conversation action.

The modeler specifies the chosen protocol (facet protocol), list of participants (facet to), communication act (facet performative) and message (facet contents).

species Initiator skills: [fipa] {
	reflex send_propose_message when: (time = 1) {
		do start_conversation to: [p] protocol: 'fipa-propose' performative: 'propose' contents: ['Go swimming?'] ;
	}

Receive messages

Each agent (with the fipa skill) is provided with several "mailbox" attributes filtering the various received messages by communication act: e.g. proposes contains the list of the received messages with the "Propose" communication act.

Receiving a message consists thus in looking at each message from the mailbox, and acting in accordance with its contents, participants...

Important remark: once the contents field of a received message has been read, it is removed from all the lists it appears in.

species Initiator skills: [fipa] {	
	reflex read_accept_proposals when: !(empty(accept_proposals)) {
		write name + ' receives accept_proposal messages';
		loop i over: accept_proposals {
			write 'accept_proposal message with content: ' + string(i.contents);
		}
	}
}

species Participant skills: [fipa] {
	reflex accept_proposal when: !(empty(proposes)) {
		message proposalFromInitiator <- proposes at 0;
		
		do accept_proposal message: proposalFromInitiator contents: ['OK! It \'s hot today!'] ;
	}
}

Remark:

  • To test that the agent has received a new message is simply done by testing whether the dedicated mailing box contains messages.
  • To get a message, the modeler can either loop over the message list to get all the messages or get a message by its index in the message box.

Reply to a received message

Given the message it has received, an agent can reply using the appropriate Communication Act (using the appropriate action). It simply has to specify the message to which it replies and the content of the reply.

Note that it does not need to specify the receiver as it is contained in the message.

species Participant skills: [fipa] {
	reflex accept_proposal when: !(empty(proposes)) {
		message proposalFromInitiator <- proposes at 0;
		
		do accept_proposal message: proposalFromInitiator contents: ['OK! It \'s hot today!'] ;
	}
}

End a conversation

When a conversation is made in the scope of an Interaction Protocol, it is ended automatically when the last Communicative Act has been sent.

In the case of a 'no-protocol conversation', it is the responsibility of the modeler to explicitly send the end_conversation message to mark the end of a conversation/interaction protocol.

When a conversation ends, it is automatically removed from the list conversations.

The message type

The agents' mailbox is defined as a list of messages. Each message is a GAML object of type message. An exhaustive description of this type is provided in the dedicated GAML Data Types page.

A message object is defined by a set of several fields, such as:

  • contents (type unknown): the content of the message
  • sender (type unknown): the sender of the message. In the case where the sender is an agent, it is possible to get the corresponding agent with agent(m.sender) (where m is the considered message).
  • unread (type bool): specify whether the message has been read.
  • emission_timestamp (type int)
  • recention_timestamp (type int)

The conversation data type

The agents' conversations contain the list of the conversations in which the agent takes part. Each conversation is a GAML object of type conversation that contains the list of messages exchanged, the protocol, initiator... An exhaustive description of this type is provided in the dedicated GAML Data Types page.

A conversation object is defined by a set of several fields, such as:

  • messages (type = list of messages): the list of messages that compose this conversation
  • protocol (type = string): the name of the protocol followed by the conversation
  • initiator (type = agent): the agent that has initiated this conversation
  • participants (type = list of agents): the list of agents that participate in this conversation
  • ended (type = bool): whether this conversation has ended or not
  1. What's new (Changelog)
  1. Installation and Launching
    1. Installation
    2. Launching GAMA
    3. Updating GAMA
    4. Installing Plugins
  2. Workspace, Projects and Models
    1. Navigating in the Workspace
    2. Changing Workspace
    3. Importing Models
  3. Editing Models
    1. GAML Editor (Generalities)
    2. GAML Editor Tools
    3. Validation of Models
  4. Running Experiments
    1. Launching Experiments
    2. Experiments User interface
    3. Controls of experiments
    4. Parameters view
    5. Inspectors and monitors
    6. Displays
    7. Batch Specific UI
    8. Errors View
  5. Running Headless
    1. Headless Batch
    2. Headless Server
    3. Headless Legacy
  6. Preferences
  7. Troubleshooting
  1. Introduction
    1. Start with GAML
    2. Organization of a Model
    3. Basic programming concepts in GAML
  2. Manipulate basic Species
  3. Global Species
    1. Regular Species
    2. Defining Actions and Behaviors
    3. Interaction between Agents
    4. Attaching Skills
    5. Inheritance
  4. Defining Advanced Species
    1. Grid Species
    2. Graph Species
    3. Mirror Species
    4. Multi-Level Architecture
  5. Defining GUI Experiment
    1. Defining Parameters
    2. Defining Displays Generalities
    3. Defining 3D Displays
    4. Defining Charts
    5. Defining Monitors and Inspectors
    6. Defining Export files
    7. Defining User Interaction
  6. Exploring Models
    1. Run Several Simulations
    2. Batch Experiments
    3. Exploration Methods
  7. Optimizing Models
    1. Runtime Concepts
    2. Analyzing code performance
    3. Optimizing Models
  8. Multi-Paradigm Modeling
    1. Control Architecture
    2. Defining Differential Equations
  1. Manipulate OSM Data
  2. Cleaning OSM Data
  3. Diffusion
  4. Using Database
  5. Using FIPA ACL
  6. Using BDI with BEN
  7. Using Driving Skill
  8. Manipulate dates
  9. Manipulate lights
  10. Using comodel
  11. Save and restore Simulations
  12. Using network
  13. Headless mode
  14. Using Headless
  15. Writing Unit Tests
  16. Ensure model's reproducibility
  17. Going further with extensions
    1. Calling R
    2. Using Graphical Editor
    3. Using Git from GAMA
  1. Built-in Species
  2. Built-in Skills
  3. Built-in Architecture
  4. Statements
  5. Data Type
  6. File Type
  7. Expressions
    1. Literals
    2. Units and Constants
    3. Pseudo Variables
    4. Variables And Attributes
    5. Operators [A-A]
    6. Operators [B-C]
    7. Operators [D-H]
    8. Operators [I-M]
    9. Operators [N-R]
    10. Operators [S-Z]
  8. Exhaustive list of GAMA Keywords
  1. Installing the GIT version
  2. Developing Extensions
    1. Developing Plugins
    2. Developing Skills
    3. Developing Statements
    4. Developing Operators
    5. Developing Types
    6. Developing Species
    7. Developing Control Architectures
    8. Index of annotations
  3. Introduction to GAMA Java API
    1. Architecture of GAMA
    2. IScope
  4. Using GAMA flags
  5. Creating a release of GAMA
  6. Documentation generation

  1. Predator Prey
  2. Road Traffic
  3. 3D Tutorial
  4. Incremental Model
  5. Luneray's flu
  6. BDI Agents

  1. Team
  2. Projects using GAMA
  3. Scientific References
  4. Training Sessions

Resources

  1. Videos
  2. Conferences
  3. Code Examples
  4. Pedagogical materials
Clone this wiki locally