Unreal Tournament 2004
PC Games Developer: Digital Extremes Publisher: Atari Rated: "M" for Mature
Strategy Guide written by JPaterson

INTRODUCTION

The ultimate sport of kill-or-be-killed is now an explosive new experience, with mind-blowing new ways to move like lightning and annihilate enemies. Welcome, one and all, to my Unreal Tournament 2004 guide for PC. Contained in this guide is everything you need to become a master player. Youll have weapon strategies, vehicle strategies, strategies for all ten game modes, as well as details on the various mutators available, and much, much more. Unreal Tournament 2004 has one of the most user friendly server browsers available. From just one screen, you can join any type of game, chat in IRC servers, and read up on the latest news about the game, including patch releases, competitions and contests. Note that this guide covers only officially released mutators and maps. Any maps made by people outside of Epic Games will not be detailed. The only time these will be detailed is if those maps are used by Epic Games in a future download. In order to install and play Unreal Tournament 2004, you will need the following system requirements (these are the minimum required): Operating System Processor Memory Hard Disk Space CD-ROM Drive Speed Video Sounds DirectX Modem Windows 98/Me/2000/XP/Linux Pentium III 1.0GHz or AMD Athlon 1.0GHz or faster 128MB RAM 5.5GB (5,500MB) 8X (DVD-ROM required for DVD version of game) 32MB Windows 98/Me/2000/XP-compatible video card Windows 98/Me/2000/XP-compatible sound card Version 9.0b (included) 33.6 Kbps modem for LAN/Internet play
If you would like more information about Unreal Tournament 2004, be sure to check out the official website at http://www.unrealtournament.com/.

PLAYER CONTROLS

These are the default controls for the game. You can change them to whatever you wish through the Game Options screen, available from the Settings link on the main menu.
Movement Controls Control Left Mouse Button Right Mouse Button Up Arrow/W Down Arrow/S Left Arrow/A Right Arrow/D Spacebar Shift Delete Page Down End What it does Fire Weapon Alternate Fire Move Forward Move Backward Strafe Left Strafe Right Jump Crouch/Down Look Up Look Down Center View

Miscellaneous Controls

Control M N + E Pause F9 Escape F1 F2 F3 F11 F12 What it does Show Path to Red Base Show Path to Blue Base Decrease HUD Size Increase HUD Size Use/Action/Activate/Get In Vehicle Pause Game (single player only) Take Screenshot in.BMP format Open Game Menu/Map Open Scoreboard Open In-Game Voice Chat Show Personal Stats Show Music Playlist Toggle Radar Map On/Off
Communication/Taunt Controls
Control T R V F ; J K L What it does Say (to everyone) Team Say (teammates only) Show Speech Menu Activate Microphone Point (animation) Pelvic Thrust (animation) Ass Smack (animation) Throat Cut (animation) Screenshots are saved to the /ScreenShots directory of wherever you installed the game.

Weapon Controls

Control Mousewheel Down/Numpad + Mousewheel Up/Numpad B G What it does Next Weapon Previous Weapon Switch to Last Weapon Switch to Best Weapon

WEAPONS - LIGHTNING GUN

The Lightning Gun is a high-power energy rifle capable of ablating even the heaviest carapace armor. Acquisition of a target at long range requires a steady hand, but the anti-jitter effect of the optical system reduces the weapon's learning curve significantly. Once the target has been acquired, the operator depresses the trigger, painting a proton 'patch' on the target. Milliseconds later the rifle emits a high voltage arc of electricity, which seeks out the charge differential and annihilates the target. -- Official description

The Lightning Gun made its debut in Unreal Tournament 2003, replacing the Sniper Rifle, and for good reason; it can only be used as a long-range weapon. The weapon is useless in close quarters fighting, as the recharge rate is very slow, and most enemies wont stand around for you to get a clean shot. The primary firing mode is an arc of electricity that shoots straight out. If aiming at the body, it will take two shots to kill an unarmored foe, three or four to kill one with armor (depending on how high the players armor is). However, since its a long-distance weapon, you should go for head shots, by aiming directly at the players cranium. A head shot will result in an instant death, regardless of the players health or armor. You also get a cool Head shot! from the announcer. The alternate firing mode brings up the Lightning Guns scope. You can zoom in a great distance to get a clear shot. Unless you have superhuman eyesight, the scope is essential for getting head shots. Keep in mind, however, that when you are zoomed in, you cannot see immediately around you, so an enemy could be doing a little dance behind you before finally deciding to kill you and youd have no idea. Besides the slow recharge rate, your view when looking through the scope is partially obstructed due to the snow you get, which looks similar to a bad connection on your television or computer monitor. It doesnt hinder your view as much as the smoke from the Sniper Rifle, but its enough to make you pay attention. One good strategy, which I used to do all the time in Starsiege: Tribes, is to get the Lightning Gun, find a good area to snipe from, and grab a vehicle, preferably a Raptor, although a Manta may also work. You use the Raptor to fly to the destination, then park it while youre sniping at your enemies. When you are spotted (or if, depending on good you are), you can jump back into the Raptor, and retreat to safety, or to another sniping location. This tactic worked wonders in Tribes, as I would use the Scout flyer to get around, then use it to run away. I call this a hit & run. Whatever you do, please dont camp near a respawn point and kill people after they spawn from a death. This is not only a cheap way to get kills, but it ruins the game for everybody. How do you improve your skill if you kill people before they have a chance to do anything? If you are, unfortunately, on the receiving end of a spawn camper, you can hold down your move forward button (default Up Arrow/W), and hold it down as you spawn. This gives you a chance to run as soon as you appear, possibly avoiding the sniper.

VEHICLES - LEVIATHAN

Originally designed for urban pacification, the Leviathan found its greatest deployment near the end of the Human/Skaarj wars. The Leviathan represents the ultimate in mobile military force projection, able to carry one driver and four passengers. Each passenger mans an anti-vehicular turret at the corners. The pilot can target enemies with the rear-mounted rocket pack while driving, firing a continuous swarm of projectiles. When the pilot needs to hold a position, the main weapon can be deployed. This will immobilize the Leviathan for stabilization. Powered by two quantum-fusion impulse reactors, the main gun projectile creates a negative singularity at the point of impact, drawing all energy and matter out of the immediate area. Exploding outward with devastating force, the shockwave can annihilate everything within its radius. There are documented cases of Leviathans single-handedly leveling entire cities and lone attacks against a fully manned Leviathan are suicide. -- Official description

Driver Controls

Control Mouse Up Arrow/W Down Arrow/S Left Arrow/A Right Arrow/D Left Mouse Button Right Mouse Button
What it does Aim Turret/Ion Accelerate Brake/Reverse Turn Left Turn Right Fire Rockets/Fire Ion Deploy/Pack Up Vehicle What it does Aim Turrets Fire Turret Zoom In/Out

Gunner Controls

Control Mouse Left Mouse Button Right Mouse Button
The Leviathan is the most powerful and most armored vehicle in the game, but all this comes at the cost of speed. An enemy runs faster than the Leviathan drives at top speed. In addition to the driver, the Leviathan can transport four players, all of whom control side turrets. While driving, the Leviathan driver has access to the rear rocket turret, which can fire a continuous stream of rockets. They home in on targets, but they arent too accurate. Each additional player has access to a side turret which fires plasma bolts. A fully armed Leviathan can be devastating, and with 6000 points of armor, it takes a lot to bring it down. The Leviathan can be deployed into an immobile mode. In this mode, the Leviathan cannot move and is more vulnerable to damage, but the driver has access to an extremely powerful ion weapon, similar to the Ion Cannon. Firing this weapon at a target creates an ion beam, and seconds later, a huge ball of ion at the target erupts in a devastating explosion, killing everyone, regardless of health or armor, and causing severe damage to any vehicles or power nodes/power core. Because it is stationary in this mode, the driver should make sure there is at least one or two players outside, ready to heal the vehicle with the Link Gun whenever it gets damaged.

CREDITS

Strategy Guide written by JPaterson. Layout and Design by Peter Ekstrom and Chris O'Rorke, Cheat Happens. Adobe version adapted by Peter Ekstrom. Screenshots captured and provided by JPaterson. Some graphics and wallpapers provided by GameWallpapers.com. Some material used for researching and writing this guide were obtained from the Epic Games website for Unreal Tournament 2004. Cheat Happens, it's owners or JPaterson are in no way affiliated with Epic Games or any of it's subsidiaries. This guide is not official or licensed in any way. This guide is Cheat Happens and may not be reprinted or reproduced, in part or in whole, in any form, including, but not limited to, electronic transmission and magazine publication, unless prior consent is given by the publisher. Unreal Tournament 2004 interactive copyright 2004 Epic Games.

Controlling Unreal Tournament 2004 Bots with the Logic-based Action Language G OLOG
Stefan Jacobs and Alexander Ferrein and Gerhard Lakemeyer
Knowledge-based System Groups Computer Science Department RWTH Aachen 52056 Aachen, Germany {sjacobs,ferrein,gerhard}@cs.rwth-aachen.de Abstract
Computer games and the accompanying entertainment industry branch has become a major market factor. AI techniques are successfully applied to tasks like path planning or intelligent swarm behavior. On the decision-making level the state of the art are state machines with a xed set of behaviors. The perception of the computer player is perfect. They exactly know where the other players are located, even if they cannot see them. This approach seems to be limited for intelligent decision making. Therefore, we propose another approach. We use a variant of the logic-based action language G OLOG for implementing so-called game bots in the U NREAL T OUR NAMENT 2004 environment. First results show that we can compete with the omniscience game bots in the Unreal domain.

Introduction

Computer games and the accompanying entertainment industry branch has become a major market factor. With increasing processor speed and powerful graphic engines computer games are becoming more and more realistic. An important aspect for game-play is that the computer opponents act in an intelligent and realistic way. AI techniques are applied to so-called low-level tasks like path planning or intelligent swarm behavior. On this level computer animated creatures seem to act human-like. A very successful branch of computer games are the socalled ego-shooter games, where the player is controlling one character perceiving the environment from the characters point of view. In those games, the characters act mainly in an adversarial environment where the goal is to eliminate opponent players. Examples for such games are Quake (Id Software Inc. 2002) or Unreal Tournament (Epic Games Inc. 2004b). Computer controlled characters are called Game bots or bots, for short. While ego-shooter games may not be everybodys cup of tea, they are nevertheless very challenging from the AI perspective. The environment is highly-dynamic, one has to deal with uncertainty, and decisions have to be made in real-time. Programming those bots in a way that the human player regards them as intelligent and intriguing opponents is a challenging task for game AI.
Copyright c 2005, American Association for Articial Intelligence (www.aaai.org). All rights reserved.
The state of the art techniques for implementing game bots are mainly state automata with xed transitions. To let them appear more realistic, the programmer equips them with a nearly perfect world model. The human player can see through this omniscience very easily. To let the behavior of a game bot appear more realistic, it should be equipped with a human-like world model. Then, the decision making algorithms become more complex, and, for AI research, more interesting. Therefore, our approach to decision making of a game bot is based on the logic-programming language G OLOG which combines explicit agent programming with the possibility to reason about actions and change. In particular, we use our variant R EADY L OG which, moreover, integrates decisiontheoretic planning techniques. Our application domain is the game Unreal Tournament 2004 which will be outlined in the next section. Then, we give an overview of the language features of R EADY L OG which we use to model Unreal game bots, before we describe the contents of the demonstration.

Unreal Tournament 2004

U NREAL T OURNAMENT 2004 (Epic Games Inc. 2004b) was developed by E PIC G AMES (Epic Games Inc. 2004a). It is a multi-player game where human players compete in disabling opponent players. The opponents are either human players or computer-controlled game bots. There exist different game modes like death match or capture the ag. The goal of the former is to disable as many opponents as possible, while in the latter two teams of players have to defend their own home base. Here, strategic planning and coordination is needed. While the game engine of U NREAL T OURNAMENT 2004 is not publicly available, the decision making routines of the game bots are open source. The bots are implemented in the object-oriented language US CRIPT, a script language which comes with the game. The algorithms for path planning or collision detection are part of the game engine and therefore cannot be modied. The bot framework in US CRIPT offers high-level actions like move to, attack, or retreat. The original bots were implemented using a state machine consisting of nine different states. The state classication is controlled by the game engine.

As the decision making code of the bots is available, it is possible to change it in such a way that agent technologies become applicable. We therefore implemented a new interface to the game engine which allows to transmit world model information like the topology of a game level, visual information of opponents, team-mates, or items like weapons or health packs. That means that these items are only transmitted if the game bot can perceive them. Using this information we build a world model which is used for specifying the bots behavior in the R EADY L OG framework. In contrast to omniscient agent programs, the information provided by the world model is incomplete in general. Previously, Kaminka et al. (Kaminka et al. 2002) used the the Unreal framework as research subject. They proposed an interface to connect to the Unreal engine which is different from ours. Mainly their work focused on the low-level tasks like path planning.

Demonstration

In the demonstration we show the current state of the game bot implementation. For the game variant death match we can show that the R EADY L OG-bot performs at an 80 % level compared to the original Unreal bots. We nd this remarkable because the world model of the Unreal bot is nearly perfect compared to our world model. For the variants capture the ag more strategic planning is needed. We are currently working on that matter. Especially in this variant of the game we expect to perform much better in the future. We want to present our framework as an agent programming framework for highly-dynamic and adversarial realtime domains. Other decision making algorithms could be easily integrated as our framework provides a well structured world model. Several recorded demonstration session can be found at http://robocup.rwth-aachen.de/readybot/.

Readylog

For specifying our high-level control we use a variant of the logic-based high-level agent programming language G OLOG (Levesque et al. 1997). G OLOG is a language based on the situation calculus (Reiter 2001). Over the recent years many extensions like dealing with concurrency, exogenous and sensing action, a continuous changing world and probabilistic projections (simulation) (Giacomo, Lesperance, & Levesque 2000; Grosskreutz & Lakemeyer 2000; 2003) made G OLOG an expressive robot programming language. We further integrated a planning module into G OLOG which chooses the best action to perform by solving a Markov Decision Process (MDP) (we refer to (Puterman 1994) for reading on MDP and to (Boutilier et al. 2000) on integrating MDPs into G OLOG). In the following we give an overview of some of the features of R EADY L OG not going into details. We successfully apply R EADY L OG in the robotic soccer domain and refer to (Ferrein, Fritz, & Lakemeyer 2005; Dylla, Ferrein, & Lakemeyer 2003) for details.

Sequence: a1 , a2 Nondeterministic Choice: a1 ; a2 Solve an MDP: solve(p, h), where p is a G OLOG program and h is the horizon up to which the MDP is solved Test: ?(c) Event-Interrupt: waitFor (c) If-then-else: if (c, a1 , a2 ) While-loops: while(c, a1 ) Condition-bounded execution: withCtrl (c, a1 ) Concurrent actions: pconc(a1 , a2 ) Probabilistic actions: prob(valprob , a1 , a2 ) Probabilistic (ofine) projection: pproj (c, a1 ) Procedures: proc(name(parameters), body)

References

Boutilier, C.; Reiter, R.; Soutchanski, M.; and Thrun, S. 2000. Decision-Theoretic, High-Level Agent Programming in the Situation Calculus. In Workshop on Decision-Theoretic Planning, Seventh International Conference on Principles of Knowledge Representation and Reasoning (KR2000). Dylla, F.; Ferrein, A.; and Lakemeyer, G. 2003. Specifying multirobot coordination in ICPGolog from simulation towards real robots. In Proc. of the Workshop on Issues in Designing Physical Agents for Dynamic Real-Time Environments: World modeling, planning, learning, and communicating (IJCAI 03). Epic Games Inc. 2004a. http://www.epicgames.com/. Epic Games Inc. 2004b. Unreal tournament 2004. http:// www.unrealtournament.com/. Ferrein, A.; Fritz, C.; and Lakemeyer, G. 2005. Using golog for deliberation and team coordination in robotic soccer. KI (1). Giacomo, G. D.; Lesperance, Y.; and Levesque, H. J. 2000. Congolog, a concurrent programming language based on the situation calculus. Articial Intelligence 121(1-2):109169. Grosskreutz, H., and Lakemeyer, G. 2000. cc-Golog: Towards More Realistic Logic-Based Robot Controllers. In AAAI-00. Grosskreutz, H., and Lakemeyer, G. 2003. Probabilistic complex actions in golog. Fundamenta Informaticae 57(1):167192. Id Software Inc. 2002. Quake 3 arena. http://www.idsoftware.com/games/quake/quake3-arena/. Kaminka, G. A.; Veloso, M. M.; Schaffer, S.; Sollitto, C.; Adobbati, R.; Marshall, A. N.; Scholder, A.; and Tejada, S. 2002. Game Bots: A Flexible Test Bed for Multiagent Research. Communications of the ACM 45(2):4345. e Levesque, H.; Reiter, R.; Lesp rance, Y.; Lin, F.; and Scherl, R. 1997. GOLOG: A Logic Programming Language for Dynamic Domains. Journal of Logic Programming 31:5984. Puterman, M. 1994. Markov Decision Processes: Discrete Dynamic Programming. New York: Wiley. Reiter, R. 2001. Knowledge in Action. Logical Foundations for Specifying and Implementing Dynamical Systems. MIT Press.
These are the ingredients for implementing successful a game bot in the R EADY L OG framework. For space reasons we cannot give an example of a game bot program.