MARIOnET: Motion Acquisition for Robots through Iterative Online Evaluative Training (2010)
Although machine learning has improved the rate and accuracy at which robots are able to learn, there still exist tasks for which humans can improve performance significantly faster and more robustly than computers. While some ongoing work considers the role of human reinforcement in intelligent algorithms, the burden of learning is often placed solely on the computer. These approaches neglect the expressive capabilities of humans, especially regarding our ability to quickly refine motor skills. In this paper, we propose a general framework for Motion Acquisition for Robots through Iterative Online Evaluative Training (MARIOnET). Our novel paradigm centers around a human in a motion-capture laboratory that ``puppets'' a robot in real-time. This mechanism allows for rapid motion development for different robots, with a training process that provides a natural human interface and requires no technical knowledge. Fully implemented and tested on two robotic platforms (one quadruped and one biped), this paper demonstrates that MARIOnET is a viable way to directly transfer human motion skills to robots.
In Ninth International Conference on Autonomous Agents and Multiagent Systems - Agents Learning Interactively from Human Teachers Workshop (AAMAS - ALIHT), May 2010.

Michael Quinlan Formerly affiliated Research Scientist mquinlan [at] cs utexas edu
Adam Setapen Masters Alumni asetapen [at] cs utexas edu
Peter Stone Faculty pstone [at] cs utexas edu