School of Computing Seminar with Avinash Ranganath, Clemson University School of Computing
Low Dimensional Motor Skill Learning Using Co-Activation
We propose an approach for motor-skill learning of highly articulated characters based on the systematic exploration of low-dimensional joint co-activation spaces. Through analyzing human motion, we first show that the dimensionality of many motion tasks is much smaller than the full DOFs of the character. Indeed, joint motion appears organized across DOFs, with multiple joints moving together and working in synchrony. We exploit such redundancy in character control by extracting task-specific joint co-activations from human recorded motion, capturing synchronized patterns of simultaneous joint movements that effectively reduce the control space across the DOFs. By learning how to excite such co-activations using deep reinforcement learning, we are able to train human-like controllers using only a small number of dimensions. We demonstrate our approach on a range of motor tasks and show its flexibility against a variety of reward functions, from minimalistic rewards that simply follow the center-of-mass of a reference trajectory to carefully shaped ones that fully track reference characters. In all cases, by learning a 10-dimensional controller on a full 28 DOF character, we reproduce high-fidelity locomotion even in the presence of sparse reward functions.
Avinash Ranganath works as a Research Scientist within the Visual Computing group, under Dr. Victor Zordan and Dr. Ioannis Karamouzas, currently working on applying Deep Reinforcement Learning for Character Control. He completed his doctoral studies from University Carlos III of Madrid, Madrid, Spain, under the supervision of Prof. Luis Moreno, in robot locomotion. His research interests include Machine Learning Character Control, Multi-robot Systems, and Locomotion.
Advisors: Dr. Victor Zordan and Dr. Ioannis Karamouzas
Friday, October 18 at 2:30pm
McAdams Hall, 114
821 McMillan Rd., Clemson, SC 29634, USA