Multi-Representational Learning from Demonstration through Sequential User Study Development
Abstract
The performer s work contributes a novel multi-representational learning framework for robot Learning from Demonstration that seeks to improve robustness, usability and performance over past algorithms. This research will leverage RobotsFor.Me, a web-based user study framework previously developed by the PI, to introduce sequential user study development with real users in complex environments. The outcome of this research will be a unified learning system that enables a robot to learn novel tasks through demonstration, guidance, correction and reward-based feedback. At the conclusion of this project we will have the ability to place a robot in a complex office-like environment, equip the robot only with primitive skills (e.g., goto, pickup), and over a period of several days enable untrained users to define and teach the high level behaviors needed for operation in that domain.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 30, 2016
- Source ID
- N000141612844
Entities
People
- Sonia Chernova
Organizations
- Georgia Tech Research Corporation
- Office of Naval Research
- United States Navy