Thriving on Chaos
Issue: Volume: 28 Issue: 3 (March 2005)

Thriving on Chaos

While most big-name computer game franchises are plagued by seemingly interminable gaps between releases, Ubisoft, creator of the hugely popular Tom Clancy’s Splinter Cell series, has been the only developer to deliver a standard-breaking sequel on nearly a yearly basis.

Only 18 months after the original Splinter Cell took the computer game scene by storm, Splinter Cell: Pandora Tomorrow followed quickly in its wake. Then, when Ubisoft announced that the next game in the series-sporting advanced AI and a completely renovated graphics and physics engine-would arrive a mere 12 months later, gamers and industry experts alike voiced their skepticism that this goal could be accomplished within such a brief development cycle. Nevertheless, for the legions of players loading up Splinter Cell: Chaos Theory for the Xbox and PC on March 22, those fears will be quickly laid to rest by cutting-edge imagery and gameplay that dramatically surpass those found in the game’s predecessors.

To ensure that Chaos Theory would attain a strong foothold at the leading edge of technology, Ubisoft charged 17 programmers with pushing every technological boundary in an effort to redefine gaming’s state of the art. Their breakthroughs included a new weather system that dynamically updates environmental and atmospheric effects, a new lighting model that merges the material and light systems for more realistic and efficient real-time lighting, and new dynamically distributed navigation meshes for more adaptive and competitive AI. By far, their biggest achievement in this first-person shooter was the application of advanced normal mapping technologies, which gave newfound detail to every object in the game, from hero Sam Fisher’s layers of Kevlar clothing to the grooves in the weaponry’s gunmetal.

Set in 2008, Chaos Theory begins when Japan forms an Information Self Defense Force (I-SDF), which is deemed by many to be a violation of both international law and its own constitution. Intent on inciting a massive world war in the Pacific Rim, the leader of the I-SDF secretly launches a misinformation campaign, igniting growing tensions among Japan, China, and North Korea. Soon, the US becomes involved, and it is up to Fisher, the National Security Agency’s most elite black-ops agent, to investigate and eliminate a new source of information attacks or risk the possibility of another world war.
In Ubisoft’s Splinter Cell: Chaos Theory, black-ops agent Sam Fisher returns for a third mission. This time, he is better trained for close combat, thanks to finely tuned “proximity” animations t

To create secret agent Fisher, as well as the countless guards, sentries, and other non-player characters (NPCs) populating the sprawling Asian game setting, the Ubisoft artists chose not to recycle the geometry from the previous titles. Instead, they built new models from scratch that are far more detailed and higher in resolution. Using Discreet’s 3ds max, the group created the model meshes, and used the software’s advanced surfacing tools to generate the various “hard” models, including battle gear (helmets, Kevlar body armor, night-vision goggles, fiber-optic cameras, and more) and Fisher’s arsenal of weapons, including a new combat knife.

As the lead character, Fisher commanded the most attention. His model has two skins, comprising 4000 and 3700 polygons, while the generic soldiers have a single level of detail (LOD) containing 2500 polygons. For those and nearly all the other models, rather than use multiple LODs, the artists employed memory-efficient normal mapping, for which Ubisoft built a brand-new graphics engine to process the added information. While a more traditional bump map encodes two planes of information (height and width), a normal map contains three vectors, or channels, of information-height, width, and length. The normal map, created from a high-res model, is then applied to a low-res counterpart. Consequently, the low-polygon model can be rendered more crisply and with greater geometric complexity than with bump maps alone, and it can be lit more realistically, casting shadows and highlights along three axes.

In contrast, all the characters in Splinter Cell and Pandora Tomorrow were merely vertex-lit and bump-mapped. “Because the [Chaos Theory] characters were already modeled in high definition, the addition of the normal maps pushed their realism beyond anything the players were anticipating,” says lead artist Pascal Beaulieu. “Sam Fisher’s face uses the same texture size as the one in Splinter Cell and Pandora Tomorrow, but it is easy to see that with the normal maps his facial anatomy and lighting is more realistic.”

According to assistant technical director Jeff Giles, the normal maps were the key to achieving the game’s realistic textures, which were made using Adobe’s Photoshop and an assortment of in-house plug-ins. “All the objects in the game are normal-mapped, which makes them look worn, weathered,” he adds. When a normal-mapped object is lit and rendered, the first render pass calculates the shadow effects of the vertices comprising the geometry, and the second render pass calculates the shadows cast by the normals of each pixel in the normal map. This is done at the texture rather than the vertex level, so it includes all the intricate, three-dimensional relief features that would be impossible, and too memory-intensive, to create with geometry.
The artists turned to normal-mapping techniques to create new, more detailed versions of Splinter Cell‘s returning cast, including hero Sam Fisher. The images from left to right show the model’s polygo

To set the characters in motion, lead animator Jacques Dussault and his team constructed two skeletons in Discreet’s Character Studio 4.0: one for Fisher and another for the NPCs, each comprising a total of 55 bones. Another 16 bones drove the facial expressions along with a proprietary phoneme system for lip sync. In addition, Ubisoft’s technical directors also wrote a host of scripts to further enhance the animation capabilities of Character Studio. For example, to achieve a natural deformation during pronation, or the rolling of an arm, the artists employed a combination of scripts and roll bones on both the forearm and upper-arm sections of both skeletons. To modify the already rigged meshes of the NPCs without having to redo the rigging and weighting, the animators used a proprietary tool that saves the skin-weight-vertex information of the skin modifier, and then reapplies it to the modified mesh.

In addition to normal mapping, Chaos Theory also sports a new real-time weather subsystem. Dynamically updating the environment as the game is played, the system can, for example, make rock walls slick with rain and have puddles form and then ripple and shimmer in the light. If a player enters a zone where it is raining, the engine dynamically modifies the specular highlights of Fisher, the NPCs, and the environmental surfaces to give them a wet, shiny look. Moreover, the system dynamically updates the normal maps to create raindrop effects on surfaces.

Along with rain, fog, snow, and lightning, the weather system can also generate wind that can affect the NPCs, the environment, and Fisher. For instance, as a storm builds and the rain thickens and a gale begins to blow, the objects in the environment-plants, smoke, characters-become wet and struggle against the wind.

Water effects also have a strategic role in the gameplay, for example, when Fisher activates courtyard sprinklers to distract guards blocking his path. To generate the plethora of fluid effects in the environments, the development group used a combination of simple particle effects, animated meshes, and normal maps. “When the sprinkler system is activated, it creates a ‘show’ effect using a combination of particles and meshes created in 3ds max,” Giles explains. “To make the area appear wet, we generated a puddle on the floor, which we gradually make visible by playing with alpha values and increasing the specularity of the puddle’s surface and any target objects.” For an object caught in the spray, the artists created beading and cascading droplets of water by adding animated normal maps to the object’s texture map and then modifying its specularity. In some instances, the artists also animated an object’s texture map to forge the illusion of water streaming across the surface.
The weather system can impact gameplay by affecting the environments and their objects, and because the system is also tied to the game’s sound system, the player can use weather-related noises to cover the sound of Fisher’s movements.

In addition, the team generated sand, dust, smoke, and other particle effects in 3ds max, all of which are produced in-game through the engine. To make the environment’s trees, grass, shrubs, bushes, and ivy respond realistically to forces such as wind and human and vehicle contact, the group rigged the foliage with a set of bones normally reserved for character animation. Then, when the foliage is disturbed, these bones are directly manipulated at run-time to create the movements.

Naturally, one of the principle attractions in any first-person shooter is the array of weapons at the player’s disposal, all of which demanded realistic particle-based imagery and volumetric effects for fire, combustive exhaust, tracer fire, and so forth. To this end, Chaos Theory marks the debut of per-pixel normal-mapped particle effects, which bombard the player with smoke, explosions, and other volumetric effects that unfold with unprecedented realism.

After previous Splinter Cell titles pushed light maps and vertex lighting to the limit, the Ubisoft team deemed it essential to generate a completely new lighting model that would mark a quantum leap forward in technology. As a result, Chaos Theory boasts some of the most impressive real-time lighting effects ever seen in a computer game, from flickering candles behind translucent rice paper doors and solid floodlights from a lighthouse, to fire, lightning, and arcing flares that dynamically reflect off wet and dry surfaces.

While light maps and vertex lighting were still used for some low-level illumination and distant scenery in the new game, the team achieved the new breakthrough in real-time lighting by merging the game’s light and material systems. In the first Splinter Cell, the lighting model was almost completely independent from the material system, with the exception of the shadows, whereas in Chaos Theory, everything is merged, making the lighting far more problematic. To resolve these issues, the artists spent three weeks tuning material characteristics, such as the per-pixel attenuation equations, to perfect the lighting properties of the textures and to remove the plastic look that often plagues normal-mapped surfaces in games.

Furthermore, the game’s new lighting engine necessitated some complicated lighting-material setups, especially for the weather system, involving a complicated shader with two normal maps. The first normal map was prepared for the material itself; the second was procedural, representing the thin layer from the water accumulation and the raindrop effects. “Considering that everything is wet, you need full light-reflection effects and a fairly strong cube map-a type of texture that projects an image from all around an object, reflecting, for example, the sky above, trees and buildings from the sides, and dirt or grass from the floor,” explains lead programmer Dany Lepage. “Then, when you add material transition (dry to wet), you get what could be the most complex material simulation ever done in a video game.”

New technological advances such as these have resulted in a massive amount of on-screen graphic content, which has to be rendered in real time. To manage this workload, Ubisoft developed a rendering engine that is almost four times as efficient as the one used for the original title. It features dynamic loading of textures, meshes, and sound, which make a huge difference in the amount of content that a level can support. “We did a lot of things more intelligently than before, including zoning most objects for prioritizing their rendering according to their proximity to the camera,” says Lepage. “This enabled us to reduce the CPU load significantly.” As a result, the engine is capable of drawing approximately 300,000 triangles per frame, which is close to 100,000 more than the norm.

This powerful rendering engine, coupled with the newly merged lighting and material system, also enabled greater transparency in the environment, significantly improving the quality of the stealth systems used for Fisher’s night-vision goggles, fiber-optic cameras, and thermal-imaging equipment. “We designed special shaders to am-plify the rendering of dark areas to increase the usefulness of night vision during gameplay,” says Lepage.

Like the rendering, Chaos Theory’s AI also surpasses that of its predecessors. After studying the original Splinter Cell, Ubisoft decided to improve on several key aspects of the game’s AI. The group’s first concern was making the player feel like the NPCs were trying harder to survive; therefore, the NPCs now run and hide when threatened, or receive proper cover fire from a teammate. Second, the team enhanc-ed the NPCs’ animation and facial expressivity to convey their emotional state. The last big step addressed the communication among the NPCs, either when they’re talking in a group or coordinating their searching and fighting. The developer laid the foundation for these advancements by completely overhauling the way the NPCs make decisions and navigate a level. By using dynamically distributed navigation meshes, the NPCs can find the best path at all times and, at a minimal CPU cost, constantly recalculate their decisions based on the state of the environment.

The NPCs also show their intelligence in other ways, such as exhibiting an awareness of their environment. For example, when rain begins to fall, the characters that are not engaged in battle will turn their palm upward, feel for drops, and then seek shelter to avoid the oncoming downpour. In addition, the AI personalizes the reactions of the NPCs to various threats, and allows them to recall past events for more intelligent behavior.
Because the normal maps didn’t extend beyond the character meshes, the artists had to perform added modeling work within 3ds max so that the silhouettes, such as this one of Fisher behind a rice paper screen, manifested a level of detail in tune wit

With this third installment in the Splinter Cell series, secret agent Sam Fisher has secured his place in the current pantheon of computer game franchises, alongside Halo’s Master Chief and Half-Life’s Gordon Freeman. In fact, the game’s characters and environments stand as the most impressive display of normal mapping yet seen in a computer game, pushing the performance of the new texturing technology beyond the standards set recently by The Chronicles of Riddick and Halo 2.

And, if Ubisoft is able to continue upgrading the series at the current pace, a new standard-bearing Splinter Cell title could arrive within a year. Right now, however, Chaos Theory is not only the most eagerly anticipated game of 2005, but it is one of the all-time high watermarks for gaming graphics.

Martin McEachern, a contributing editor for Computer Graphics World, can be reached at

Sam Fisher has returned with greater flexibility and acrobatic agility, which he can employ in a variety of stealth combat maneuvers-including climbing, crawling, hanging from his hands or legs, and swinging. To accommodate these actions, the artists developed a three-step process for constraining all the characters’ feet to the ground so they’d move without sliding.

The first stage uses inverse kinematics to calcu-late the distance required to make the foot carrying Fisher’s weight touch the ground. The leg is then bent at the hip, knee, and ankle, while the other leg’s position is fixed in the world coordinate system until the weight is transferred to the other foot. If the distance between the stepping foot and the ground is too great, the system lowers the hip until the feet touch the ground. Similarly, the hip bones are adjusted for moving uphill or downhill.

In the second step, Alias’s HumanIK (formerly from Kaydara) calculates the torsion and constraints applied to each bone, namely the toe, ankle, knee, hip, and spine. This is done based on assigned parameters and the weighting and design of the actual skeleton.

In the final step, the system analyzes the data compiled by HumanIK, and if the torsion is too high or the bend in the leg is too acute, the system gently reverts to the original animation, putting the weight on the other leg or both feet. The engine then renders the result before returning to the initial step to repeat the process for the next animation frame.

In Chaos Theory, Ubisoft introduces a new multiplayer, cooperative mode, in which two players can coordinate their maneuvers on missions playable over the Internet or a LAN. For instance, when charged with infiltrating a glass-sided building that is fronted by a large wall, one player can crouch down, clasp his hands together, and boost his teammate over the wall and onto a pipe that leads to the roof of the building. Once on the roof, the character can lower a rope for his partner to climb, and even motion to his teammate to pull the rope left or right, thereby moving him out of a guard’s view. When they’re inside the building, the players can peer under doors using a fiber-optic camera, and can share the view through each other’s camera. Finally, faced with hacking a computer next to a sleeping guard, a character, perched inside an air duct, can lower his teammate on a rope, dangling him headfirst over the keyboard, Mission Impossible style.

According to lead animator Jacques Dussault, the greatest challenge in creating these cooperative animations involved connecting two characters with some element of the environment or an NPC. “For example, lowering your teammate by a rope to neutralize an unsuspecting NPC below required the connection of three characters, a deceptively complex mixture of animation and programming,” he notes. “This involved figuring out the starting and ending position of the animations for a character moving in space and coming in contact with another, which is tedious work that requires a good deal of patience.”

After reviewing the original Splinter Cell, Ubisoft found that the most empowering feeling for the player was when Fisher was extremely close to an enemy and in control of the situation. Therefore, of particular importance were the animations that focused on proximity.

When Fisher and a guard converge on each other, their respective animations blend into a series of “close proximity” animations, such as a crouching, furtive walk. For instance, when Fisher gets close to an NPC, his head suddenly turns toward the guard and his steps become more cautious and calculated. In a more threatening situation, the game engine automatically activates an animation in which Fisher, for example, prepares to grab the NPC by the neck.

The team also developed a number of unique animation cycles for Fisher centered on a new weapon: a combat knife. And, the animators reworked all the standard animations-walks, grabs, attacks-to make them less acrobatic and more thoughtful and purposeful than those from previous games.

To string these movements together, the team used a blending system that employs three types of blends. The first, and most commonly used, is “tweening,” which transitions between two animations as soon as the current one is interrupted. The second type of blend combines two or more animations, and the resulting animation, for example, could be 40 percent of a walk and 60 percent of a jog. The third type of blend, called additive blending, counterbalances the displacement of one animation with another, and is used primarily for facial animations.

While light maps and vertex lighting are still used for low-level lighting and distant scenery in Chaos Theory, nearly everything else is lit with per-pixel omni lights and per-pixel spotlights, which can cast fil