GDCE Day 3: Conference Day 2

Keynote Panel: Playing All Platforms - Convergent games

Friday got off to a rather frustrating start, as I got caught in traffic and missed the fast train into London and as a result, took an extra half an hour to get in, causing me to miss the first part of this talk. Chaired by Ian Livingstone of Eidos, the panel consisted of Mickey Kalifa of Playjam - OpenTV, Rio Caraeff of Sony Pictures Mobile, and Heikki Aura from Nokia.

Arriving late, I only caught Mickey Kalifa's presentation about the OpenTV platform. This particular platform has proven surprisingly lucrative in the Europe, and is soon to open in the US. Played through the television through Sky, this platform works on a pay per play basis. The games themselves are relatively simple - platformers and puzzle games predominantly, and the games are changed every six weeks to keep the players interested. While the shelf life of these games is very short, so is the development time, no more than six to eight weeks in total.

State of the Art in AI and Design

Chaired by Demis Hassabis of Elixir, this presentation took the format of a number of case studies, covering AI solutions for various different platforms. First up was Rob Davis of Solaris Media, the developers of the educational strategy game Battlefield Academy, developed for the BBC.

Battlefield Academy is a turn based strategy game developed with Flash. The limitations of the Flash platform are fairly obvious: it's an interpreted language, and it's delivered over the web, meaning speed and size are both limited.

In order to make the most of the available CPU resources for the AI, Solaris Media kept the graphics and animations down to a bare minimum. The performance restraints were so tight that they could only implement a very simple A* for pathfinding. In order to reduce the search space they kept the visibility for each unit low and carefully designed each map to prevent too many pathfinding problems.

The developers made good use of the map file for embedded AI hints, such as the locations of nearby objectives, which saves CPU cycles while enabling units to respond sensibly to nearby threats and goals. They also managed to squeeze in a very simple form of influence mapping, which uses a single pass to get a rough idea of the shape of the battlefield and the position of the front line.

The Battlefield Academy game is available online from here:

http://www.bbc.co.uk/history/war/launch_gms_bfacademy.shtml

The next speaker was Pierre Pontevia of Kynogon, a middleware developer working on AI for integration with Renderware engines.

AI can be broken down into three basic steps: Perception, Decision, and Action. Perception here is key, since without suitable information about the world, the AI cannot make good decisions. Traditionally, AI's have relied on simple 'workarounds' to perception and decision-making, such as ray casting techniques and scripting. As game environments get larger and more sophisticated these methods will prove to be inadequate.

Pontevia demonstrated their dynamic 3D topology analysis tools, which generate a model of the game world that the AI system can use to determine areas of interest. They showed off some impressive demos, such as the 'Bodyguard' demo - three body guards surround the player and reposition themselves intelligently to cover areas from which a threat might present itself. More information and demos are available from Kynogon's website: http://www.kynogon.com/

The final talk was presented by Hendrik Heimer of Navus, a middleware company specializing in MMORPG AI. Heimer started by highlighting the advantages and disadvantages of the scripts and FSMs used in many current games. On the plus side, they are relatively easy to implement, and give the designers a high degree of control over the NPCs and thus, by extension, the storyline. The minuses are rather obvious however; NPCs exhibit robotic and uninteresting behaviour, poor or non-existent interaction both with the player and with each other, and they are weak outside a manually tuned environment.

Heimer identified the requirements of the next generation AI as being: Rich communication amongst NPCs and between NPCs and players, Machine learning, and 'freedom of mind' for NPCs - that is, removing as many artificial limitations from their behaviour as possible. More information about their Multiagent system and MMORPG AI backbone can be found at their website, http://www.navus.de.

The Easy Route To Console Online

Presented by Shawn Hargreaves of Climax, this session suggested some tips and tricks to getting a good multiplayer experience with a console game.

Hargreaves identified fan communities as being an invaluable resource - given enough tools these communities will happily develop new game modes and features that you didn't have time to develop yourself - or else were unable to develop due to licensing issues. As an example of this was the 'Demolition Darby' game mode of Moto Cross, which would never have been allowed by the license, and yet is one of the most popular game modes online.

He also suggested of carefully tailoring your scoreboards to a wide range of players. Rather than have a single 'top ten' or anything like that, have a series of tables so that players of all levels can feel they have accomplished something. Also be wary that cheating can easily ruin a basic high score table. He mentioned a 'slipstream pausing' technique used in Moto Cross, which is rather tedious to pull off, but gives you ridiculously short lap times that can only be beaten by someone else using the same cheat. As a result, making the top ten is no longer an exciting challenge, but a very boring exercise.

Another item he felt was of key importance to online gaming was the user interface. All too often the user interface for setting up multiplayer games is farmed out to the least experienced programmer, and as a result, it is often full of bugs and poorly designed.

Mods, Mission Packs, Doom 3 and Beyond

Perhaps the most entertaining session of the day was given by Splash Damage, the developers of the Wolfenstein: Enemy Territory expansion pack. In this talk they gave a post-mortem of the production of W:ET, and talked about how they went from an amateur mod development team and evolved it into an award winning development company.

What went Right:

Plenty of reference material: A trip to a war museum gave the developers a chance to actually play with the weapons used in the game, allowing them to make them highly realistic, without needing to create large amounts of concept art or making stuff up.

The Interface: Although they admitted that it's rather ugly and complicated, the Splash Damage felt that the interface of the game really worked - by having everything on one page players can set up and configure their character very quickly in game, without the need to navigate lots of menus.

The Engine: The Quake 3 engine is well known, and there are plenty of resources available for developing mods with it. This made the learning curve much easier, and meant that the team had a solid platform to work with.

Level Design: Each map in W:ET is a team effort, rather than the work of one developer. Each aspect of map design is given to the team member best suited for that particular work, ensuring that each map is very high quality. They also managed to avoid wasting too much time on dud maps - they work with a two pass map creation process - the first pass consists of the basic geometry, and playing it to make sure it's actually fun, the second pass consists of making it look nice.

The Game Design Document: The game design document was well maintained throughout the project, undergoing fifty-eight revisions over the course of development. Splash Damage shied away from posting it on the intranet in order to allow everyone on the team to modify it, preferring instead to let the designers come up with the base ideas and hold regular team meetings to get feedback.

Throwing Out Rubbish: Every so often, something included in the game would turn out to be rubbish. Splash Damage were strict on throwing this stuff out, even if it meant missing a milestone. This didn't always go down too well with the publisher, of course, and the team stressed the value of having a good relationship with a heavyweight developer such as Id - if Id agreed that a feature was no good and should be thrown out, then Activision would defer to their judgement.

What Went Wrong

Not Enough Artists: With only two artists on their team, Splash Damage had a hard time generating the amount of content required for the game, and with their next project using the Doom3 engine, this is likely to be even more problematic due to the number and quality of the textures needed to show off the engine at it's best.

The HUD is 'Pants': Splash Damage admitted that the head up display in Enemy Territory is utterly horrible, with nearly every corner of the screen being used for some bit of pop out information - half of which the players don't even know about.

ECTS