Here are the results for the 2nd Annual AiGameDev.com Awards for Game AI, where the best games of the year are nominated and voted by professionals, enthusiasts, and researchers in artificial intelligence for games.
Best AI in a Mainstream Game
Left 4 Dead
Left 4 Dead obliterated the competition, the AI in the game standing out for multiple reasons:
As with previous Valve games, there is strong NPC interaction with comrades reacting to a massive amount of circumstances, shouting to comrades and having their own banter between themselves.
The NPC AI for the human survivors is able to help assist the PC up until the hardest difficulty, and easily take over at a moments notice for an idle player in online multiplayer. The individual zombie AI is good too, with various special zombies all being able to get to survivors wherever they are on a level, causing havoc.
The “AI director”, which works to spawn enemies and items, makes playthroughs different each time.
Here’s what readers and voters had to say:
“I think it’s obvious that Left 4 Dead should be up there. AI characters behave in believable, competent manners… the zombies have fantastic pathing AI, the bosses can be surprisingly devious, and the AI director does a fantastic job of regulating the experience.” — Ian Morrison
Left 4 Dead got the majority of the votes, with the finalists all having less, but more equal amount of votes:
Finalists: Fallout 3, Grand Theft Auto 4, Resistance 2, S.T.A.L.K.E.R.: Clear Sky
Technical Innovation in Game AI
Left 4 Dead
Left 4 Dead won a close race for Technical Innovation in Game AI, coming up against some reasonable competition from EA’s Spore. It stands out on its own as an impressive feat of technical work however with:
An impressive set of spawning and path finding technologies that allow the well-animated zombies (who look impressive when cornering!) to run anywhere in a level, chasing down survivors wherever they go.
The “AI director”, collecting together many technologies to allow the pacing of the level to change by spawning items and zombies, and doing zombie attacks at different times - providing a different playthrough each time.
Here’s what readers had to say:
“in my book they deserve an award just for *attempting* an AI director, to say nothing of really pulling it off.” — Kevin Dill
Spore was hot on the heels of Left 4 Dead, with its own achievements in procedural animation well observed.
Runner Up: Spore
Finalists: Fable 2, Far Cry 2, Star Wars: The Force Unleashed
Best AI in an Independent Game
Sins of a Solar Empire
Sins of a Solar Empire won out against the competition with it’s competitive multiplayer AI:
The AI is able to control masses of ships and plan expansion between planets, as well as retreating when in a losing battle and planning attacks against weak points in opposing empires.
Fleet AI is very good, allowing players to usually send their own strategic orders to attack an enemy planet and not have to worry about individual fleet battle tactics.
Defcon’s new AI came a close second this year, with Be the Dinosaur and Multiwinia being the other finalists.
Runner Up: Defcon
Finalists: Be the Dinosaur, Multiwinia
Most Influential Research Publication 2008
Automatically-generated Convex Region Decomposition for Real-time Spatial Agent Navigation in Virtual Worlds
The winning paper takes navigation mesh generation up to the next level, being able to automatically generate it for any given level space. From the paper:
This paper presents a new method for decomposing environments of complex geometry into a navigation mesh represented by bounding geometry and a connectivity graph for real-time agent usage in virtual worlds. This is accomplished by the generation of a well-de?ned and high-coverage set of convex navigable regions and the connected gateways between them. The focus of this paper is a new automated algorithm developed for decomposing a 2D representation of world-space into arbitrary sided high-order polygons. The DEACCON (Decomposition of Environments for the Creation of Convex-region Navigation-meshes) algorithm works by seeding a 2D polygonal representation of world-space with a series of quads. Each quad is then provided with the opportunity to grow to its maximum extent before encountering an obstruction. DEACCON implements an automatic subdividing system to convert quads into higher-order polygons while still maintaining the convex property. This allows for the generation of navigation meshes with high degrees of coverage while still allowing the use of large navigation regions, providing for much easier agent navigation in virtual worlds. Compared to the Space-?lling Volumes and HertelMelhorn navigation mesh decomposition methods, DEACCON provides more complete coverage, controllable mesh sizes, and better overall algorithmic control to desired decomposition quality with an improvement in agent navigation speed due to better decompositions.
Automatically-generated Convex Region Decomposition for Real-time Spatial Agent Navigation in Virtual Worlds shows how better results can be obtained when building navigation meshes.
Automatically-generated Convex Region Decomposition for Real-time Spatial Agent Navigation in Virtual Worlds Authors: D. Hunter Hale, G. Michael Youngblood, and Priyesh N. Dixit Download PDF
Runner Up: Real-Time Planning for Parameterized Human Motion
Finalists: A Cover-Based Approach to Multi-Agent Moving Target Pursuit, Planning with Hierarchical Task Networks in Video Games, The Rise of Potential Fields in Real Time Strategy Bots
What do you think about the results? Who would have you like to see win an award? Click here to post a comment!