impress.js | presentation tool based on the power of CSS3 transforms and transitions in modern browsers

Project Outline

Created a team of behavioural Agents designed to compete in a game of Capture the Flag (CTF)
Agents are programmed in the agent programming language IndiGolog
The Entitys that the Agents control exist in the UnrealEngine, an advanced 3D game engine
Created an implementation of the Environment Interface Standard (EIS) to connect the two different environments
Integrated research on CTF game strategy

CTF Rules and Strategy

CTF is a team-based game
Goal: Steal opponents’ flag and return it to the friendly team’s base
Combat tactics are required to prevent being attacked while carrying the enemy’s flag back to the friendly base
Team roles:
were created to improve the effectiveness and believability of the AI team:

Capturer	Goal is to capture the enemy teams flag
Capturer Support (Capsup)	Goal is to assist the Capturer in its goal
Defender	Goal is to prevent the friendly team’s flag from being taken
Long Range Support(LRS)	Goal is to support the meta-game, influencing team mates and supporting team-mates

Two strategies
were developed to evaluate the performance of the IndiGolog Agent team:

Diversion	The Capsup and LRS attract the enemy’s attention so the Capturer can reach the enemy flag with as little resistance as possible
Fast Attack	The Capsup and the Capturer go directly to the flag, not diverting to attack enemies and with less regard for stealth. The number of attempts made to capture the flag is increased.

Agent Design

Situation Calculus

A language used to represent dynamically changing worlds
IndiGolog and Situational Calculus actions are specified in recursive form
- <actionName>(<value>, ... ,<value>, Situation) where the Situation is the sequence of actions that occurred after the Initial Situation.
Example:
- startgoto(enemyflag, assignrole(jim, capturer, assignname(jim, S₀)))
The state of the Agent’s knowledge base comes from its knowledge in the Initial Situation plus the sequence of actions after.
In the above situation:
- All of the initial conditions according to S0 are true except for
- The name of the agent is jim
- The role of the agent with name jim is capturer

Agent Design

Successor State Axiom

Agent Design

IndiGolog Implementation

The abstract model of the environment the Agent uses does not depend on the 3D environment the Entity exists in
The Agents are defined by a set of fluents (relations that vary from situation to situation) and its main procedure in IndiGolog
The abstract model of the environment the Agent uses does not depend on the 3D environment the Entity exists in
Example procedure:
- some([agent], and(name=agent, and(role(agent)=capturer, hasItem(enemyflag,agent) ))),[startgoto(homeflag)]
If there is an agent with the same name as this agents name and that has the item “enemyflag” then start going towards the homeflag
The fluents in the above procedure are
- name, the agents assigned name
- role(<Payer>), the role for the <Player>
- hasItem(<Item>, <Player>), true if the <Player> has the Item <Item>
startgoto(<Location>) is a primitive action
Procedures consists of other procedures and/or primitive actions

High and Low Level Controllers

To make the Agents more effective two kinds of controllers are used
Low level controller
- Designed to abstract out all of the tedious parts of executing actions such as knowing the exact location of the Entity and planning paths for actions
High level controller
- Works with an abstract model of the environment
- Receives action information about environment from low level controller
Example: Movement
- The low level controller will handle knowledge of the location of the Entity and planning the optimal path to take to the destination
- The high level controller will decide the destination with respect to the state of its abstract model of the environment

High and Low Level Controllers

Implementation of the EIS

A customized version of the EIS was implemented for this project
The Agent and Entity Classes were expanded upon to include data on their relationships
The EIS functions as a messaging server and data converter between the IndiGolog environment and the UnrealEngine
Each Agent has a list of associated Entities and a connection to the Agent in another environment

When a message is received to the Agent's connection it is sent to all of the Entities in the Agent's list of associatedEntities (similar for Entities)

Findings

Synchronization is important in a multi-agent environment
- Due to asynchronous aspects of the UnrealEngine and the TCP/IP network, team members would be out of sink with each other
Agent programming languages are great tools for modelling agent behaviour
- They can form a much cleaner design
- Debugging is made simpler
- Actions are chosen with respect to large knowledge base

High and low level controller design
- Separation of high level behaviour design and low level action sequence scripting has allowed for a more believable agent
- Low level interface improves usability of high level controller
- Issue with where to draw the line between what should be handled by which controller
Future Work
- Planning could be used to make more informed action choices
- Interfaces for low level controllers could be made easier to develop

For the following videos the Blue team is the team of IndiGolog Bots and the Red team is a team of the default Unreal Tournoment Bots that come with the UDK.

This movie demonstrates the successful use of a diversion strategy used by the blue team.

This video demonstrates the successful execution of the fast attack strategy.

Thank you