Abstract:

We describe how to automatically synthesize motion controllers for locomotive tasks involving animated characters modeled as 3D mass-spring lattices. The motion controllers determine an actuation sequence based on elapsed time, not physical state; actuation is represented economically using a canonical set of global lattice deformations; and stochastic search is used to determine effective values for the controller parameters. Our algorithm generates controllers that produce stylistic, visually plausible motion for simple locomotive tasks in under an hour on a standard workstation, which is more than an order of magnitude faster than comparable approaches to motion synthesis for 3D articulated-linkage models. Key Words: Spacetime constraints, controller synthesis, stochastic optimization.