Neuroscience and Robotics Laboratory, Northwestern University
Robotic Marionettes: Bridging the Dynamics Gap
This talk will describe a project aimed at creating a robotic system and associated software infrastructure that, together, are capable of autonomous performance of marionette shows using classical humanoid marionettes. This is a difficult problem due to the marionette’s high level of possible articulations driven with a limited level of control authority. In order to provide expressive, human-like motions one must learn to understand and leverage the complex, nonlinear dynamics inherent to this system. I will discuss a variety of techniques that we have used to automatically synthesize control trajectories for generating dynamically-feasible motions, and I will discuss several methods used to derive mathematical descriptions of the desired motions necessary for the synthesis techniques. I will include high-level descriptions of these techniques, discussions of their effectiveness and applicability, and several demonstrations.
BIO: Jarvis Schultz has collaborated with Disney Research to develop custom hardware and software platforms for embedded control of robotic marionettes using choreography and human- motion capture data. His talk will address issues in design, planning, and control for real time and pre-recorded motion sequences. In particular, he considers how marionettes (which are actuated by strings) make use of abstraction and force parametric animation to generate dynamic robot performers.