The manual construction of controllers involves hand crafting the control functions for a set of muscles. This is often possible when the muscle activation patterns in the corresponding real animal are well known. For example, in Miller's Miller88 work, the crawling motion of the snake is achieved via a sinusoidal activation (i.e. contraction) function of successive muscle pairs along the body of the snake. Another manually constructed controller for deformable models is that for controlling facial muscles [Terzopoulos and Waters1990, Lee, Terzopoulos and Waters1995] for realistic human facial animation, where a ``facial action coding system'' controller coordinates the actions of the major facial muscles to produce meaningful expressions. Most manually constructed controllers have been developed for rigid, articulated figures: Wilhelms Wilhelms87 developed ``Virya'' -- one of the earliest human figure animation system that incorporates forward and inverse dynamic simulation; Raibert Raibert91 showed how useful parameterized controllers of hoppers, kangaroos, bipeds, and quadrupeds can be achieved by decomposing the problem into a set of manually-manageable control problems; Hodgins et al. Hodgins95 used similar techniques to animate a variety of human motions associated with athletics; McKenna et al. Mckenna90 produced a dynamic simulation of a walking cockroach controlled by sets of coupled oscillators; Brooks Brooks91 achieved similar results for a six-legged physical robot; Stewart and Cremer Stewart92 created a dynamic simulation of a biped walking by defining a finite-state machine that adds and removes constraint equations. A good survey of these sorts of approaches can be found in the book by Badler, Barsky and Zeltzer Badler91.
Fish animation poses control challenges characteristic of highly deformable, muscular bodies, not unlike those of snakes [Miller1988]. We have devised a motor control system that achieves muscle-based, hydrodynamic locomotion by simulating the dynamic interactions between the artificial fish's deformable body and its aquatic environment. To derive the muscle control functions for fish locomotion, we have consulted the literature on marine biomechanics [Webb1989, Blake1983, Alexander1992]. The resulting parameterized controllers harness the hydrodynamic forces on fins to achieve forward locomotion over a range of speeds, to execute turns, and to alter body roll, pitch, and yaw so that the fish can move freely within its 3D virtual world.
The main drawback with manually constructed controllers is that they can be extremely difficult and tedious to derive, especially for many-degree-of-freedom body motions. Moreover, the resulting controllers may not be readily transferable to different models or systems; nevertheless, they can serve as a good starting point for the optimization-based algorithms.
|Xiaoyuan Tu||January 1996|