For robots to be effective assistants in a wide variety of everyday environments working alongside humans they must
be capable of adapting to new settings and learning new skills and tasks as needed with minimal overhead. This project explores the application of crowdsourcing to the problem of interactive task learning with the aim of enabling remotely-located users to effectively teach robots to perform complex tasks in real-world environments.
We developed a novel system that allows users to demonstrate hierarchical tasks via web-based control of a table-mounted six degree of freedom robot arm. The system employs intelligent action grouping suggestions and substitution suggestions for error recovery to assist the user in providing quality demonstrations.
We are currently investigating the effects of different recruitment methods, the use of suggestions and other active querying during demonstrations, and also combining input from multiple users during a single demonstration in several human-robot interaction scenarios.