week 12 [ May 2sd -> May 9th]
Purpose: [Artificial player prototype] starting the design of the bot
- starting writing the report about the prototype results. A first version
will be provided on Friday.
- starting the design of the final program.
week 12 [April 27th -> May 2sd]
Purpose: [Artificial player prototype] analyzing prototype results
Test: the test program consists in a loop. An event is provided
to initialize the process (event = data of the Defender and attacker
aircraft at the beginning). The event is computed by the AI module. A
command is provided. Basic actions are given to the model that
updates the positions. A new event is created and provided again to the AI
module. The loop is ended. Now only 30 loops are executed.
To visualize the results: the events are logged and plotted. The
following graphs show how they change:
- TURN RATE: to be satisfactory, the Attacker turnRate should
remain as close possible as the defender turnRate.
- SPEED: to be satisfactory, the Attacker speed should remain as
close possible as the defender speed.
- ASPECT ANGLE: to be satisfactory, the aspect Angle should remain close
to 0.
- RANGE: to be satisfactory, the aspect Angle should remain as lo as
possible
-
FIRST CONCLUSIONS not so satisfactory they show that there are
still some implementation errors. No more can be said for the
moment.
week 12 [April 21th -> April 27th]
Purpose: [Artificial player prototype] prototype achieved
The prototype consists of 3 parts:
- AI layer: it takes 'events' as inputs and computes 'commands' as
outputs. [implementation almost achieved, it has to be tested]
- Model layer: it updates positions and simulates the aircraft
behavior. [implementation achieved, it has to be tested]
- Intermediate layer: it processes the 'commands' to create 'basic
actions'. It uses some standard trajectories as data to provide feasible
'basic actions' [not yet implemented]
NB: the *.ppt file is available here: click
here
week 11 [April 14th -> April 21th]
- In the prototype: providing the basic maneuvers has been almost achieved.
- In the prototype: detecting the incompatible basic maneuvers is achieved.
week 10 [April 7th -> April 14th]
Purpose: [Artificial player prototype] prototype achieved
- The prototype decision trees have been implemented.
- PROBLEM: how to detect that decisions provided
('pursuit', 'acceleration', 'turn') are not incompatible?
- Meeting conclusion: some further details have been given. it would be interesting
to illustrate the behavior of the basic decision-making process by
simulating a very basic dogfight situation: given an enemy's trajectory, it
would be nice to visualize the attacker automated behavior to position. A
kind of basic aircraft model is required to update the aircraft
positions.
week 9[March 31st -> April 7th]
Purpose: [Artificial player prototype] achieving a basic decision tree for
the prototype, starting the prototype design
- The prototype implements a very basic functionality of the dogfight bot.
One of both aircraft is a defender and the other one the attacker. The
prototype is a program that computes the attacker's decisions such as 'pursuit'
(the way the attacker points at the enemy), 'acceleration', 'turn'
(how tight is he turn) according to parameters as both aircraft relative 'positions',
'speeds', 'headings'. Nevertheless the information as the
enemy's trajectory, energy considerations have no been taken into account in
this first step.
- The decision process is based on decision trees. To decide which kind of
pursuit the attacker will perform, a decision tree has been worked out. The
leaves give information about the pursuit to be performed. Two other
decision trees on 'acceleration' and 'turn' have been worked out.
- The decision trees are available (AccelerationTree.jpg, PursuitTree.jpg,
TurnRateTree.jpg). Click
here
- Some inputs and corresponding outputs have been given as examples
(example1.jpg, example2.jpg, example3.jpg). Click
here
NB: the original *.ppt file that gathers all this information is also
available: click
here
Week 8 [March 24th -> March 31
st]
Purpose: [Artificial Player prototype] prototype
implemented, project definitive proposal updated
- the proposal was written down. It is available : click
here
- a schedule has been provided in the proposal.
Week 7 [March 17th -> March 24 th]
Purpose: [Artificial Player prototype] start coding
and start writing the proposal
Week 6 [March 10th -> March 17 th]
Purpose: [Artificial player prototype] designing the
decision making process of the dogfighter prototype: more precisely having a
decision tree for the dogfighter prototype.
- The decision tree has not yet been finished. For the moment, it has been
suspended. The first priority is writing the proposal for the end of the
week 7.
Week 5
Purpose: [Artificial player prototype] write a document that explains how a
real pilot chooses a maneuver and how the pilot model could make its choice.
- Introduction to the bases of dogfight
- Description: the making decision process of a real pilot: why does he
choose this maneuver? what is taken into account (positional geometry, energy, weapon envelop…).
- Description : how a model can take data into account to be as close as
possible to a real pilot behavior.
- Problems: lack of information about maneuver
knowledge, some data must be confidential. solution suggested: maneuver will
be ideal curves.
- March 11 th, 2003, release of report that sums up these ideas: ChooseManeuverReport
Week3-4
Purpose: implement a basic Artificial player prototype
- The initiative to implement a basic Artificial player prototype has
been taken. The goals are the following:
- have a precise ideas about the limits that will be encountered in the
pilot modeling process (lack of information)
- deal with Artificial Intelligence issues
- be used as soon as possible to Pilot environment vocabulary
- be able to write a realistic proposal within the end of March
- The artificial player prototype will be very simple: it will model a
dogfight pilot: only the attacker will be developed. Many simplifications
have been made (few maneuvers will be implemented).
- A lot of information have been gathered about dogfight maneuvers: here are
the documents and urls:
Week 3
Purpose: border the project topic :"implementing a Flight Simulator
Artificial Player"
- First approach to our Flight Simulator Artificial Player.
- First problems: many pilot specific
skills are missing to design an entire bot.
- An overview of the layered architecture has been given: what deals
with Artificial intelligence and what deals with pilot specific skills
have been clearly separated.
- Two layers : AI layer (making decision process) and know-how layer
(pilot specifics kills acquired through practise)
- It has been decided that the project will focus on the AI part.
- a short report that stands for a first proposal has been written: FirstFSBotProposal
on February 18 th , 2003. But it should be updated within the end
of March.
Week 1-2
Purpose: get as quick as possible information about the ICE project
(Intelligent Cockpit Environment)
- Reading a AI report: The Quake III Arena Bot, by J.M.P van
Waveren.
- Reading F16 report: A knowledge base describing the situations an F16
pilot might encounter, by Quint Mouthaan
- Reading docs about Flight Gear : Flight Gear Installation and Getting
Started, by Michael Basler and Martin Spott
downloadable on http://www.flightgear.org/
Updated May 5th 2003