I have a ton of new notes from MiGs this year and while I miss the debates and discussions from the first few MiGs it remains a really good venue for thinking through game development in more meaningful ways. I hope the new organizers next year will protect and develop this aspect.
One of the most interesting talks I went to was a session with A2M’s Liko-Paul Pinsonnault on day two entitled, “The art of dying: where animation meets physics.” Liko-Paul is a software developer and programmer who has some super interesting things to say about managing the techno-cultural constraints of animating NPC death sequences verses their procedural generation.
At the heart of the talk for me was the point that digital games as games can never be reduced to simulations – in a nutshell, what gets lost in the move from animation to simulation is “style.” Let’s explore…
So the story goes that the designers wanted NPC deaths in WET to look cool. Its a game inspired by Kill Bill and the like and so I gather they were after a kind of aesthetic of wholesale slaughter that wouldn’t seem monotonous or rote. Now, I haven’t played WET yet so I can not comment on whether they pull this off. Instead, I was interested in the story of how they first tried animation, found it was too labour and memory intensive and so switched to a physics simulation…. but this “looked bad” so they settled on a combo of both.
Just to be clear, we are talking about what the player sees after they have dealt a death blow/shot to an NPC combatant in the game, i.e. how the NPC dies. The trick here of course is to have a way of dying consistent with the imagined game design and play. In some games, you kill a NPC and it just disappears, or falls over but clearly the designers of WET want to have an aesthetic consequence for players’ actions but as gameplay mechanic (to signal that the NPC is dead and therefore won’t be fighting anymore) and as a reward for, you know… killing well. A well executed death blow deserves a well executed death scene.
Animation then is a stock solution for this… in effect the death blow triggers a micro-cut scene of one of several or many animations of dying. The game machine notes the dynamic state at the moment of death and then given a number of variables (bullet, sword, high, low, near wall, on ledge etc…) simply runs one animation available in a huge database. The bigger the database the better the variety and continuity will be in the way NPCs die. The problem is that if you really want stylish death and not stock death then you need a lot of animations, a lot of costly animators and a lot of memory for that database. This is even trickier when you want to have multiple (Kill Bill) style deaths in the same scene.
One memory/labour cost-effective solution to all this is to use a “rag doll” physics engine to govern the movement of a NPC body after a death blow/shot. CSI like — the engine calculates all the forces acting on the body, takes into account physical obstructions and the like and hell maybe even wind speed… whatever you like. Given that every instance of death is different in at least some of the variables then boom – there you go – instant death scene auto-generated on the fly and each one is slightly different and very believable. Layoff the animators.
Not so says Liko-Paul – each rag doll death is different but each rag doll death looks like a rag doll death — no style at all unless its rag doll you are after. So in the context of the aesthetic and story-telling requirements physics simulation doesn’t cut it. The moral of the story – you can’t reduce a game to procedurality… move over some more ludologists…
Indeed, the non procedural/non-interactive cut-scene is enjoying a resurgence – from Uncharted2 through Dragon Age to the micro-animations of WET… these interventions are a source of “style” in the otherwise predictable unpredictability of simulation. This is to say nothing of the culture clash between animators and programmers over the control of representation in games…
The A2M solution in the end was to use a bit of both… start of with an animation and then finish with a simulation. Liko-Paul showed some hilarious consequences of what happens when you shift from animation to simulation without being careful. It seems that animators routinely don’t have to care about things like gravity and velocity, mass and stuff so if you use animation vectors as inputs for a physics simulator you can get NPCs bodies that spin wildly and explode and stuff. Its a real trick to get it to look right I guess.
So there you have it — I just love the absurdity of unpacking all this through the notion of dying NPCs in games but it was a beautifully done talk, subtle and humble, with great game studies implications.