Blurred Vision: Autonomous Characters with Cinematic Camera ControlBen Resner, Bill Tomlinson, Hao Yan MIT Media Laboratory {benres, badger, yanhao}@media.mit.eduABSTRACT In this project, we describe the design and implementation of an autonomous character system withspecial cinematic camera control. A 3D autonomous character (a dog) is made to wander in a 3D scenehunting another 3D character (a rabbit). The dog is equipped with a virtual nose to locate the rabbit.Value-based behavior control methods are employed for modeling the autonomy of the characters. Amotor system with simple collision control is implemented for moving the characters, switchinganimations, and avoiding colliding with scene objects and boundaries. To achieve cinematic effect inscene rendering, a synthetic camera automatically alternates between an interesting third person view ofthe scene and first person views from the dog’s and rabbit’s perspective. We also implemented depth offield effects such that only a slice of the Z-depth is in perfect focus. Other related graphics technologieswe tackle are inverse kinematics and texture mapping. The system is implemented in Java with AGTanimation file format and runs on Windows NT. Figure 0: The rabbit being chased by the dog. 1. INTRODUCTIONIn this project, we design and implement an autonomous character system with cinematic cameracontrol.We?ve made an autonomous dog that hunts for a rabbit. A synthetic camera chooses a variety of shots toeffectively convey the events that happen during the chase. The dog and rabbit exist on a street mapdownloaded from Mapquest.com. The rabbit is constrained to follow the streets. It can make sharp turnsand move fast. The dog is not constrained by the map, but must avoid randomly placed obstacles.Simple bounding box collision is used for obstacle avoidance. Also, the dog cannot make sharp turns. When far away from the rabbit, the dog searches for scent trails the rabbit leaves behind. When the dogis proximate to the rabbit, it directly accesses the rabbit’s location and pursues in a straight line. In order to show off the interaction between the dog and the rabbit, we built an automaticcinematography system to decide on the position and orientation of the camera at every clock tick.Traditional film cinematography has a lot to offer in this respect. We tried to draw on the heritage offilm, modeling the cinematography of the scene after the speeder bike chase scene from "The Return ofthe Jedi." To focus viewers? attention of the most important element of each shot, we implemented depth of field.Depth of field is a powerful cinematic technique which adds a dimension of realism to our work.Traditional first person render viewpoints show all objects -- near and distant -- fully in focus. Withdepth of field, only a portion of the world in the camera?s view frustum will be sharply rendered.Changes in the shot show by the camera, coupled with depth of field changes, will produce a dramaticsense of shifting attention. This should help the user form an emotional connection to the characters. We’re using Java user interface controls to allow the user to modify scene parameters in real time. Theuser can control the speed of the rabbit and dog, and the dog’s turning radius. This allows users toexperiment with values to attain optimal dramatic timing between dog and rabbit. The scene and characters were modeled and animated in 3D Studio Max. The animations and geometrywere exported using the AGT exporter from Animetix Technologies. The dog model was animatedusing the "bones" system of inverse kinematics, with occasional hand touch-ups. Care had to be taken inthe model to avoid unnecessary transforms that can confuse the scene-graph.In the next section, we briefly describe the goals of this project. We then describe in detail the designand implementation of important system components. The description of our achievements and lessonslearned follow. We then list the contributions to the project by each group member. We conclude thisreport by proposing some interesting future work. 2. GOALSWhile planning this project, we had several main goals with respect to both the process and the product.We wanted to design a course of implementation that would allow each of us to learn at least one areathat we were not familiar with. We also wanted to produce a product that worked well as a unifiedwhole. We chose the dog and rabbit project because it had room for the three main areas of work we wanted topursue: autonomous behavior, animation and cinematography. On the behavior front, Hao needed tolearn how to design an autonomous system that could respond appropriately to an ever-changingenvironment. With regard to animation, Ben needed to learn how to shepherd an animation all the wayfrom 3D Studio Max until it appears on the screen in our graphics system. Bill needed to figure out howto apply his background in cinematography to the world of computer graphics, specifically choosing thedepth of field problem as one that would force him to dig all the way to the bottom of the graphicsengine.Since each of us has strengths that the others lack, we chose a project that would cause us to worktogether fairly closely, so that we could leverage each other?s skills if we got stuck. This worked outquite well, with each of us having something to bring to the table.3. SYSTEM DESIGN AND IMPLEMENTATIONIn this section, we describe the design of important system components, including Character Design,Environment Design, and Cinematic Camera Control.3.1 Character DesignHaving nice 3D character models and animations is not enough to make a compelling autonomouscharacter. To achieve the autonomy of the character, we need a mechanism that decides what to dounder certain conditions. Following [Blumberg & Galyean, 1995], we built a simple value-basedbehavior control system for our characters. A motor system, which provides the behavior system withhigh-level commands such as MoveForward, PlayAnimation, etc., is built separately such that itminimizes the amount of "house-keeping" required of the behavior system.3.2 Behavior system It is helpful if we take the "Intentional Stance" [Dennett, 1987] of the characters when design them.Thinking about a complex system by means of the intentional stance suggests that the most effectiveway to understand that system is by means of attributing intentions, drives, motivations and desires to it.All characters have certain desires
View Full Document
Unlocking...