34 July 2000/Vol. 43, No. 7 COMMUNICATIONS OF THE ACM

Transcription

1 34 July 2000/Vol. 43, No. 7 COMMUNICATIONS OF THE ACM

2 The plesures of immersive interctive entertinment incresingly demnd gme designers deliver physiclly consistent nd convincing experience solving fundmentl computtionl problems long the wy. Physics Gmes in Computer Chris Hecker SCREENSHOT FROM PAPYRUS DESIGN GROUP'S GRAND PRIX LEGENDS I don t know bout you, but when I ctch ir in my Ferrri 312 rcing down the strightwy t Monco, I look forwrd to the gyroscopic effects of the wheels nd engine rotting the cr in mid-flight. I especilly enjoy it when I crsh into wll or nother cr, flip over, nd the wreckge not to mention my stnding in the 1967 Grnd Prix circuit creens down the trck, tumbling nd bouncing s it goes. In contrst, I don t like it t ll when I cn t stck up bunch of crtes to get over the wll of the cstle in which I m being held cptive. Or, worse, it relly bugs me when I cn t even move the crtes, becuse they re stuck fst to the floor through no mens I cn discern. In the virtul world of tody s incresingly immersive nd ddictive computer gmes, the physics of the environment the very rules governing how the entities in the gme interct is subject to the whim of gme designer, the skills of tem of gme progrmmers, nd the speed of the CPU on which the gme is running. As these skills nd CPU speeds improve, nd s plyer expecttions increse, gme physics simultors hve to keep pce by becoming incresingly relistic nd dvnced. Therefore, gme developers re turning to serious lgorithms from the computer science reserch community to bring their virtul worlds to ever higher levels of sophistiction. Here, I survey the use of physics nd COMMUNICATIONS OF THE ACM July 2000/Vol. 43, No. 7 35

3 its role in the multibillion-dollr interctive entertinment industry, then prognosticte on where gme physics is likely to go in the ner nd not-soner future. Physicl Simultion in Gmes Every computer gme involving moving objects, from the simplest puzzle gmes like the clssic Tetris, even checkers, to the most incredibly relistic uto rcing or flight simultion, depends on physics simultor. Ech of them needs some core lgorithm to determine the rules under which the objects in its gme world move round. Is flling Tetris shpe blocked on one side by nother piece? Cn the checkers piece move to the left, or does it bump up ginst the edge of the bord? Cn the wheels of my rce cr exert enough force on the ground to ccelerte the cr without slipping? Wht hppens when I djust the trim on my ilerons? These queries nd lots more re nswered by the physics simultor, to vrying degrees of sophistiction, nd the results re displyed to the user. Although the physics simultor in Tetris or checkers would not be recognized s such by most cdemic simultion reserchers, it contins ll the necessry ttributes: It tkes inputs nd current positions of objects in the environment; computes the movement of the objects bsed on certin rules; nd updtes the object positions when ll interctions re resolved. Then it repets. It doesn t solve differentil equtions of motion or use F=m but does produce motions consistent with the rules of the virtul world (if you ll llow me to cll the checkers bord virtul world). The complexity of the rules of motion nd the wys interctions re resolved re the only things differentiting the Tetris simultor from its counterprt in the Grnd Prix cr rcing gme. The most importnt question for gme developer to sk is whether the physics simultor in prticulr gme is good mtch for the design. Does it produce the motions the plyer expects nd needs in order to hve fun plying the gme? Under this consistency metric, the Consistency is crucil for keeping plyers immersed in gme environments; it s oky to be ble to run 100 miles per hour nd jump 20 feet s long s the rest of the world rects ppropritely. Tetris simultor is flwless. Unfortuntely, most current gme simultors score miserbly under this metric. Even the best of the crefully tuned rcing nd flying gmes contin holes in their environments, where the gme s physicl lws ct in nonintuitive wys, confounding nd frustrting the plyer. The problem is tht when gme designers try to crete virtul world more complex nd subtle thn Tetris, the limittions of the hrdwre, softwre, nd their own brinpower quickly come into ply. No one knows how to write simultion lgorithms tht simulte the rel world, or even the slightly crtoonlike world of Indin Jones. And even if developers did know how to crete these lgorithms, it s uncler how much processing power they would consume if the world were filled with hundreds of objects. So they simplify, then simplify some more. Eventully, however, they will come up with lgorithms pproximting the minimum number of physicl rules they decide re necessry to ply their prticulr gmes nd hope they hven t introduced ny gping holes nd inconsistencies in the environment plyers might fll through. Although conventionl wisdom in the gme industry sys the gol of the gmes is relity itself, especilly s infinitely chep nd powerful computers become commonplce, the rel gol is consistency. Relity is boring; escping relity is why people ply computer gmes in the first plce. Consistency, on the other hnd, is crucil for keeping plyers immersed in gme environments; it s oky to be ble to run 100 miles per hour nd jump 20 feet s long s the rest of the world rects ppropritely. In effect, ech computer gme mkes contrct with its plyers when they strt plying; ny inconsistency violtes this contrct, reminding the plyer it s just gme. Getting stuck in wll due to bd collision detection is not something plyer wnts to hppen when dventuring in dungeon. When inconsistencies occur like when boom microphone flots into movie frme during 36 July 2000/Vol. 43, No. 7 COMMUNICATIONS OF THE ACM

4 n emotionl scene the cretors of immersive environments hve filed. Antomy of Gme Simultor Before exploring further the kinds of physicl inconsistencies plguing tody s gmes nd how the industry hndles them, I briefly outline the flow structure of modern gme, looking t the vrious stges constituting physics simultor. Computer gmes run in loop. They tke input from the plyer, figure out wht the non-plyer-controlled entities in the gme re supposed to be doing, then combine this informtion into coherent new stte. Finlly, they render the result. The physics simultor, which is centrl prt of this loop, consists of four min stges: contct detection; contct resolution; force computtion; nd stte integrtion. Some reserchers might rgue with my imprecise decomposition nd note tht lgorithms differ from simultor to simultor, but these stges do the job for this exposition. Contct detection. Contct detection is the mostly geometric tsk of figuring out where the objects in n environment re reltive to one nother. Objects cn be in ny of number of reltionships, including disjoint, penetrting, resting on ech other, nd colliding violently. The oft-herd phrse collision detection is subset of contct detection. Contct resolution. The contct-resolution stge determines wht the physics simultor hs to do bout the collisions nd other contcts detected in the contct-detection stge. Are the objects touching now but supposed to be moving prt? Cn they do both? Does the simultor need to do something to prevent penetrtion, or re pir of objects penetrting one nother nd thus need to be seprted to be consistent? Force computtion. All the forces influencing the objects re clculted during the ptly nmed forcecomputtion stge. Here, the simultor considers physicl forces like the wind resistnce on cr, the recoil from grende luncher, nd trctor bems from enemy spceships, or mundne forces like grvity pulling objects down to the ground. This stge intercts with the contct-detection stge in some simultors, becuse one wy to prevent the interpenetrtion of objects is to hve them exert forces on one nother. Stte integrtion. Finlly, the forces in the forcecomputtion stge re pplied to the objects, nd their sttes re dvnced over time. To do so, the simultor numericlly integrtes the equtions of motion highly mthemticl process. Bsiclly, stte integrtion nswers the question of how fr n object moves when force cts on it. After stte integrtion, the updted positions of the objects re redy to be rendered in their updted positions by the gme s grphics subsystem. If these new positions re slightly different from the old positions, nd the gme loop cn execute these four stges fst enough, the objects look to the plyer s though they relly re moving cross the screen. Consistency Problems If the simultion lgorithm is tht strightforwrd, where do the problems come in? Ech stge in gme s physics simultor is rife with simplifictions, ech of which cn led to consistency problems. For exmple, in order to increse the performnce of the contct-detection stge, the gme uses simple pproximting shpes for the objects, even though such objects my be rendered with more detiled representtions. These pproximtions speed up the geometric queries, even during contct detection. A humn chrcter rendered with thousnds of polygons might be pproximted in the physics simultor s sphere or bounding box. This simplifiction cn led to inconsistencies whereby it visully ppers the plyer cn squeeze through n re, though the simplified physicl representtion mkes it impossible to do so. Approximtions re lso mde in the contct-resolution stge. Becuse it s so difficult for progrmmer to compute the correct wy to keep stck of boxes upright, it s sometimes esier for the designer to not let them stck up in the first plce. It s lso importnt to be ble to put objects to sleep, so the gme doesn t spend time computing physics for objects tht ren t importnt to the gme experience t prticulr moment, though this too cn led to inconsistencies. Ply ny gme tht seeks to deliver some degree of relism, nd you ll find these inconsistencies nd worse. It could involve wys of solving problems tht simply don t work, becuse they weren t nticipted by the gme designer nd hrd-coded into the gme s solutions. Some gme genres generlly provide better physics thn others. For exmple, gmes simulting specific mchines, such s flight simultors or cr rces, hve pretty solid physics simultors delivering convincingly relistic nd consistent experience. These domin-specific gmes hve n esier time thn gmes tht try to simulte humn beings wlking round in cluttered rooms. Simulting speeding rce cr my seem more complicted thn simulting wlking humn, but it s ctully much esier, becuse it s more defined problem. There re well-known physics equtions governing the limited number of wys crs behve in motion; these wys of behving re clled the cr s degrees of freedom. Humns, by contrst, hve hundreds, if not thousnds, of degrees of free- COMMUNICATIONS OF THE ACM July 2000/Vol. 43, No. 7 37

5 dom, nd the humn brin exerts incredibly subtle control over ech one of them. Physicl Control Tht lst sentence suggests dunting problem physicl control fcing future developers of gme physics. Even though I ve complined bout the stte of ffirs of physics simultion, some types of simultor technology re getting close to the point t which they stisfy gme s demnd for consistency. I m confident tht specific simultion problems, such s those involving rigid bodies, will be solved theoreticlly by grphics nd physics reserchers in the next five yers. There my be performnce issues, but s long s gme designers re creful bout the number of objects they llow to interct t given time, lgorithms will work consistently. Rigid bodies will be ble to model most of the physicl objects gmes need for t lest the next severl yers, including the usul crtes, brrels, tbles, chirs, swords, nd other inflexible shpes. They ll lso be ble to model rticulted figures, including robots, even humns, by constrining lots of bodies together with joints (see It remins n open question how gme designers cn tell drmtic story with its nturl twists nd turns when everything in the gme is controlled by physicl lws. Popović s Controlling Relistic Chrcter Animtion in this section). Physicl control in both computer gme progrmming nd leding-edge reserch is completely unsolved problem. Control is the work of exerting forces nd torques on objects to get them to do wht the gme designer wnts them to do, like chse the plyer. Physicl control is bout knowing which forces need to be computed during the force-computtion stge. It s fine to hve robust simultor with detiled rticulted figures for ech chrcter in the gme. But which torques does the physics simultor hve to put on the joints to mke the chrcters stnd up nd wlk like flesh-nd-blood humns? A gme populted by rg dolls tht just fll over nd sit on the floor is no fun. You might be wondering how gme chrcters mnge to move round t ll in tody s gmes if physicl control is such difficult problem. Most gmes tody use cnned nimtions to control their chrcters movements. They involve very simple physicl model for ech chrcter, s in the box nd sphere pproximtions mentioned erlier, nd the physicl model moves round the world while the chrcter s grphicl representtion nimtes. The chrcter s limb movements don t ffect the physicl simultion in ny wy t ll; such prescripted nimtions re usully either hnd-nimted or motioncptured from humn performer. While hnd nimtion nd motion cpture re fine for noninterctive entertinment like movies nd television, they provide yet nother plce for inconsistencies to crop up in n interctive environment. Sy you hve n nimtion of chrcter swinging sword in hllwy. In movie, the performers cn mke sure they don t hit their swords ginst the wlls s they swing. If the ctors ccidentlly do hit the wll, the director stops the performnce nd films nother tke. In video gme, the sword-swinging nimtion hs to be creted before the plyer hs even thought bout buying the gme. Imgine tht plyer s chrcter is in the hllwy nd redy to swing the sword. Wht if the chrcter is too close to the wll becuse of where the plyer hppened to plce him before the swing? The plyer hits the swing-thesword button nd the chrcter s sword flies towrd the wll. At this point, one of two things usully hppens: The sword goes right through the wll s if it s not there, or the sword touches the wll nd the nimtion stops premturely. Neither of these options is convincing, consistent, or immersive, but when the movements re ll prescripted, the gme developer rrely hs nother choice. There re mny kludgey wys of prtilly fixing this problem, but physicl control is the only generl solution. When physiclly controlled chrcter swings his or her sword nd hits the wll, the sword collides with the wll, nd the impct force ripples bck up the chrcter s rm. It looks consistent, nd the world feels solid nd rel. The plyer s immersion is not compromised. 38 July 2000/Vol. 43, No. 7 COMMUNICATIONS OF THE ACM

6 A lot of work remins to be done in the gme industry nd in the reserch community before physicl controllers cn be leverged effectively in gmes. Gme developers usully drw from reserch ppers s soon s computers re fst enough to run the published lgorithms in rel time. For physicl controllers, however, the reserchers don t yet know how to mke humns wlk convincingly let lone run down flight of stirs while dodging bullets. Physicl control is very difficult problem too difficult for gme developers to solve in the midst of product cycle. They need to rely on control theory nd robotics reserchers to solve this problem first. In the erly 1990s, these reserchers focused on the field of physicl control for locomotion. Tody, however, fewer of them seem to still be working on this problem, lthough physicl control is more importnt thn ever, s interctive entertinment becomes n incresingly populr medium. While some of them continue in this re, I implore others to get into it nd help dvnce the stte of the rt. The Future of Gmes Over the next few yers, I expect most video gmes to completely embrce physicl control nd simultion. Interctivity the key fctor differentiting gmes from other forms of entertinment will demnd it. The interctivity nd choice of wht to do nd where to go tht plyers will incresingly hve in gmes mens the computer hs to be ble to generte consistent results from n infinite vriety of inputs. There s simply no wy to prescript or nticipte ll the options. Becuse gme developers cnnot know hed of time how close to wll plyers will be when swinging their swords, the resulting movement hs to be generted lgorithmiclly. As you cn probbly tell from wht I ve written here, the best lgorithms for generting movement re those tht obey consistent set of physicl lws. These lgorithms llow for plyers physicl intuition to step in nd help them sty immersed, spring them unplesnt surprises. However, it remins n open question how gme designers cn tell drmtic story with its nturl twists nd turns when everything in the gme is controlled by physicl lws. Every time physicl lgorithm is dded to replce prescripted sequence, the gme designer loses some rtistic control. Gme design s n rt form will hve to evolve to hndle this loss of control, just like movie directing evolved to hndle movble cmers nd other cinemtic innovtions. Mny hurdles hve to be overcome before gmes convincingly simulte physicl environments nd humn beings. In the ner term, ever-prgmtic gme developers will use bits nd pieces of physicl simultion to dd relism nd consistency to their worlds, without messing up populr nd proven gme designs. Explosions, objects tumbling, nd other esy-to-simulte effects will continue to rise the qulity of computer gmes until the dy gme chrcters re completely interctive nd physiclly rective to their worlds. c Chris Hecker (checker@d6.com) is technicl nd rt director of Definition Six, Inc., Oklnd, Clif., nd editor-t-lrge of Gme Developer Mgzine Chris Hecker COMMUNICATIONS OF THE ACM July 2000/Vol. 43, No. 7 39