Smash hits and flops: a brief history of ragdoll physics

April 26, 2016 by Douglas Fry

From noodle arms to... well, more noodle arms, ragdoll physics have been adding realism and challenge to games since the mid-90s. Here's how ragdolls got their start.

We are witnessing the fall of Ronda Rousey, mixed martial arts champion. Her head snaps backwards from a powerful kick, shifting her centre of balance. Rousey reels backwards, knees buckling as she falls unconscious. It's all over - again. Not an upset title fight defeat, but a virtual rematch in EA Sports’ UFC 2.

This is a battle of wiry skeletons padded out with polygons, their physical universe governed by algorithms. A collision was detected: forces were calculated, and they were strong enough to warrant rendering a knockout. Virtual Rousey crumples and flops in a calculated conspiracy between the simulated gravity and her biometric parameters. She is a victim of ragdoll physics, the simulation of force on articulated character models and the consequent effects on their posture and movement.

Ragdoll physics are well-suited to a combat sports title, no doubt, but also commonplace in the wider world of modern gaming. We’ve seen countless dead bodies drop in high-profile blockbusters, and stacked near-naked men in low-key indie works. In the grand scheme of videogames, however, ragdoll physics are a relatively recent phenomenon – partly a by-product of advances in computer processing power and hardware. Real-time “physically based modelling” received much academic interest in the 90s, and made an early outing in 1996 with Looking Glass Studios’ Terra Nova: Strike Force Centauri. The sci-fi shooter employed inverse kinematics (IK) in animating its characters, whose leg poses were shaped by the 3D terrain. In 1998, key members of the Terra Nova team further developed this animation style in the maligned-yet-revered Trespasser.

An ambitious first-person Jurassic Park spinoff, Trespasser aimed to holistically simulate real-world physics, where objects and character models responded dynamically to gravity and player interaction. Key to this system was the protagonist’s arm, animated using IK so that collisions, acceleration and inertia all defined the arm’s vast (and awkward) range of poses.

IK also governed dinosaur animation, with Traspasser designer Richard Wyckoff conceding in a soul-searching analysis that the system produced unpredictable results: “[T]he dinosaur legs frequently stretched, bent, and popped as the IK system struggled to handle impossible movements,” so the dinosaurs “often looked bizarre because of this lack of realistic bounds.” Even so, Trespasser’s animation system was pioneering, and the age of ragdoll physics was born with the slump of a dead Velociraptor.

In the same period, Danish mathematician Thomas Jakobsen had developed algorithms, outlined in his seminal Advanced Character Physics, which simulated physical effects on objects by calculating the movement of the particles they contained. “[S]imply move the particle positions proportionally to the force inflicted on them and the velocities will be adjusted automatically,” Jakobsen explains. Applied in gaming terms, striking a character would cause them to recoil, or topple in a particular direction. Jakobsen’s work was central to the comprehensive physics system in IO Interactive’s Hitman: Codename 47, released in 2000. Hitman featured a number of gaming firsts, including deformable cloth and plants, but its ragdoll physics enabled “a gameplay paradigm shift,” according to the game’s executive producer, Jonas Eneroth.

“It was actually the core concept and what enabled the project as a whole,” Eneroth says. “Very much a new and key differential feature that impacted story, gameplay and the overall vision.” Realistic simulation of gunfire on humans was a neat theatrical touch, but the ragdoll mechanics served a central gameplay function, allowing players to drag and conceal incriminating bodies. The game’s Glacier engine formed the basis of later IO titles like Freedom Fighters and Kane & Lynch, and carried the Hitman series through to Blood Money in 2006.

An even more widely applied physics engine emerged around the same time. During the 1990s, a research group at Trinity College Dublin developed a flexible physics simulation that was initially used in educational games, and later commercialised with a government grant. The group rebranded as Havok and showcased its software development kit at GDC in 1999.

Hugh Reynolds, Havok co-founder and early Trinity College group member, recalls it was “somehow magic” to simulate physically responsive objects. “So you've got this incredibly emotive type of animation, and yet the people who needed to get their hands on it were kinda locked out,” Reynolds says. “Physics was the purview of the ‘physics programmer’ not the storytellers. Havok was about letting the storytellers in.” The middleware allowed developers to easily implement a stable and reliable physics system, ranging from ragdoll animations to particle effects. While its commercial debut in 2000 with London Racer was underwhelming, Havok technology soon became ubiquitous, underpinning the physics in everything from the Matrix films to Valve’s vaunted Source engine.

A standout early adopter was Max Payne 2 in 2003, which gruesomely blended the series’ signature slow-motion effects with Havok’s pliable physiology. The following year, Psi-Ops: The Mindgate Conspiracy encouraged players to telekinetically toss enemies around like – well, like ragdolls – and Havok adoption by stealth games like Thief: Deadly Shadows and Splinter Cell: Chaos Theory allowed bodies to be unceremoniously dumped.

For the time being, it seemed like ragdoll physics were almost exclusively the province of death and injury, a spectacular reward for virtual murder. In 2001, however, Oxford-based software company NaturalMotion began exploring the potential for walking, talking, living ragdolls that were consistently subject to the laws of physics and biomechanics – even if they had not been felled with a headshot. With the team’s academic expertise ranging from biology to engineering, NaturalMotion developed the Euphoria engine, which “allows characters to dynamically synthesise motion in response to the events around simulating a character’s motor nervous system, body and muscles.” In other words, interaction with objects and forces determine a character’s animation in real-time.

Rockstar Studios quickly recognised Euphoria’s potential and in 2008, integrated the technology into its RAGE engine for Grand Theft Auto IV. As Niko Bellic pushed through the crowded streets of Liberty City, pedestrians would stumble out of the way, stepping back and throwing out their arms to regain their balance. Along with subsequent RAGE outings like Red Dead Redemption and Max Payne 3, Euphoria was also integrated into LucasArts’ Star Wars: The Force Unleashed.

Besides high-profile studio use, the Euphoria engine was also employed by NaturalMotion itself in 2009 to develop Backbreaker, an American football game. In a promo video, Backbreaker producer Matthew Best highlights a game sequence where players are tackling, grabbing, jostling for position, their movements shaped by Euphoria’s algorithms. Prior to physics engines, Best says, “Each interaction would be loving crafted by an animator, or would be recorded by an actor in a mocap suit. If you look at the complexity of all these interactions going on here, it’s just impossible to imagine how many animations it would take to make that look good.” In this regard, ragdoll physics serve a very utilitarian function for developers, reducing the time (and production costs) required to animate a vast range of character motions. Although Backbreaker received lukewarm reviews, it was a compelling proof-of-concept for later sports games. Consequently, dynamic animations now feature in EA Sports franchises like FIFA, Madden NFL and (of course) UFC, underpinned by its in-house Ignite engine.

Jonas Eneroth says that modern ragdoll physics in game development “is a good thing if used properly - it should not be a replacement for good animation or rigging. That is always needed.” But looking beyond development considerations, why is it that we, the audience, are so taken with ragdoll physics as a game feature? Do we derive satisfaction from seeing human bodies drop and contort in a tangible manner, behaving as we would expect in the real world? Is this a matter of projecting twisted fantasies, or augmenting the player-character bond?

Hugh Reynolds suggests that ragdoll physics “are all about empathy - when you merge this with really nicely done animation or motion capture and then transition correctly into ragdoll mode it's incredible.” Perhaps manipulating physical objects in a familiar way also allows us to better connect with a broader game world. As Thomas Jakobsen notes in Advanced Character Physics, “accuracy is not really the primary concern [in developing a physics-based game]...rather, here the important goals are believability (the programmer can cheat as much as he wants if the player still feels immersed).”

There are, of course, instances where a physics engine miscalculation ruins this immersion. Though sometimes frustrating for the audience, these glitches can produce entertaining gaming experiences in their own right, like the well-documented host of jittery encounters in Fallout 4, the unexpected acrobatics of Skate 3, or the demonic contortions of UFC2 fighters. Some players will even deliberately break and exploit game physics for the amusing results – a mod for Grand Theft Auto V, for example, randomly imposes ragdoll behaviour on character models. Developers, too, have openly embraced the silly possibilities of ragdoll systems: NaturalMotion’s mobile app Clumsy Ninja turns the Euphoria engine into a sandbox toy, while Bugbear Entertainment’s FlatOut racing titles leverage the ability to launch a ragdoll driver through the windscreen in a range of minigames, such as ten-pin bowling and darts.

Even the most ridiculous manifestations of ragdoll physics are grounded in some semblance of reality, however, and the Newtonian laws that the audience is inherently familiar with. “It's like something switches on in your head,” Hugh Reynolds says. “You ‘get it.’ I think it's a really deep instinctive thing for us all.” When we can naturally intuit the forces being simulated on game characters, we can predict the outcome of a particular action with reasonable accuracy.

This holds especially true in games where players have been granted control of characters’ individual limbs, like Gang Beasts, Octodad and Grow Home. In these games, the ragdoll characters are like marionette puppets, their arms and legs manipulated by extensions of the player’s own movements. With the advent of virtual reality, this form of player-character connection stands to be enhanced even further. Job Simulator: The 2050 Archives, for example, is as much a physics tech demo as a VR showcase, allowing players to pick up and throw in-game objects. It points to the potential for the player to manipulate ragdoll NPCs with precision gestures, like puppeteers wearing a bulky headset.

But what happens when the direction is reversed? With VR, the player is the ragdoll. When force is applied to their in-game presence, how will this impact a player’s own body? Experiencing a ragdoll collapse from a first-person perspective, while the player stands static in their living room, is potentially disorienting and immersion-breaking. And what about the buffeting effects of bullet strikes or car accidents – can we expect the integration of force-feedback, and are we comfortable with such physical effects? Until these issues are borne out, we’ll have to be content with letting virtual avatars like Ronda Rousey take the beating on our behalf.