Issue: Volume: 23 Issue: 8 (August 2000)

Designer Simulations

The ability to use physical simulations to drive animations has been a boon to animators who struggle to digitally model "difficult" motion, such as that embodied by a ball bouncing, a window breaking, water flowing, or cloth folding. In each of these scenarios, the behavior of multiple visual characteristics must be taken into account at every time step.

The downside to existing physical-simulation approaches is that they tend to be "hands-off." Users are typically not able to control or edit the simulations. This can be a particular nuisance when the simulated motion produces an undesirable visual effect. For example, the animator might want shards of glass to fall in locations different from those resulting from the simulation or he or she might want a bowling ball to follow a specific path. In such instances, the user must continually change the input parameters of the initial simulation until the desired effect is achieved, waiting until each simulation is complete before being able to assess the visual outcome. Getting the "right" look can be a long, tortuous process.
A new simulation-control technique lets animators chart the course of collision sequences by manipulating bodies. The underlying algorithm computes the required physical parameters to enable the motion.

Given the growing popularity of physical simulation in the animation world, re searchers are scurrying to develop tools that provide users with some level of interactive control over the simulations. A team at Carnegie Mellon University has taken a significant step in this direction with its development of a system that enables interactive control over rigid-body simulations. With the technique, an animator can select objects in a scene at any point of motion and drag them to the desired locations. The system then computes the required physical parameters and simulates the resulting motion. The entire simulation-editing process runs at interactive speeds, enabling the animator to quickly design complex physical animations.

"Our goal was to develop a tool that would allow physical simulation to be controlled in a more intuitive way," says principal researcher Jovan Popovic. "We wanted an animator to be able to create an animation by telling the system what the bodies should do, then leaving it to the system to figure out how to achieve that so it both looks physically realistic and the bodies do what the animator desires."

This approach enables the animator to quickly explore the range of possible motions that could lead to the desired end result. "For example, the artist might want to show a ball going through a hoop, and then might want to change the motion-maybe make the ball bounce off the backboard in a certain way and roll around the rim," says Popovic. "[With this technology], the animator can interactively guide the system through that process by grabbing and changing the state of the ball at any location along its trajectory, directing not only where the body should be, but how that effect should be accomplished-what to bounce off of, and so forth." The algorithm then computes the physical parameters that will enable the specified motion.

The interactive rigid-body simulation system comprises three components: a differential control module, a rigid-body dynamics simulator, and a user interface for motion display and editing. The first component, the animator-driven control module, is the vehicle through which control parameters are computed based on the user's movement of the mouse to specify the desired motion. The parameters are then fed into a general-purpose rigid-body simulator, which recomputes the motion and updates the display appropriately.

The motion display depicts the entire path of all the objects in the scene at each point of interaction. If there are three objects in a scene, the display shows the trajectories of all three objects. "The artist can then slide to any point in time, which effectively means moving the objects to their positions at that point in time, and can then drag the object in the desired direction," says Popovic. The system then recomputes the entire motion of the body, versus the edited portion alone, in order to maintain physical realism of the event.

The complexity of the desired simulation will affect the system's performance. "You can obtain interactive manipulation as long as the simulator quickly computes the entire motion," says Popovic. Currently, the system can handle simple motions of three or four objects, but not complex interactions among many bodies. This will change as processing power and simulation capabilities advance.

A significant advantage of the simulation-control system is that it leverages off of the uncertainties of the digital world. For example, says Popovic, "when a ball collides with the floor, the person observing that doesn't precisely know what the texture of the floor is, so we can change the texture [in the simulation], offsetting the true normal of the geometry, thus changing the end result. In this way, we are able to obtain additional degrees of freedom so the user can impose more goals on the resulting motion."

According to Popovic, the system has a way to go before it could be utilized as a general-purpose simulation editor. One of the system's main strengths-the high level of control it gives animators to chart collision sequences-can also be a drawback. "We rely on the animator to specify the scenario. For example, if the animator wants the object to bounce off the wall, then off the door before it lands into the basket, that motion sequence has to be defined. But you can imagine cases where guiding the collision through a complicated sequence would be too tedious. For example, if you were looking at the pool break in billiards, you're not going to want to specify the collision sequence of all of the balls." In an effort to avoid such tedium, the researchers are looking into the possibility of automating the collision-specification process in some situations. However, Popovic notes, "that wouldn't be an interactive process. You'd need more off-line time."

The researchers are also considering ways to extend the system to such applications as linked rigid bodies (a chain or human skeleton, for example). One way to achieve this might be through a tighter integration between the simulator and the control mechanism, which are mostly decoupled right now, says Popovic. "Linking them more closely could make the method more flexible and improve its efficiency."

Ultimately, such capabilities will find their way into the same animation authoring tools that have begun to implement physical simulations, says Popovic. "Animators will be able to create and edit complex physical animations, controlling not only the process, but also the end result."

Diana Phillips Mahoney is chief technology editor of Computer Graphics World.