User observation & dataset collection for robot training

• Published in: 2011 6th ACM/IEEE International Conference on Human-Robot Interaction (HRI) pp. 217 - 218
• Main Author: Pantofaru, C.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.03.2011
• Subjects: Computer vision; Humans; Lasers; Measurement; Robot sensing systems; Training
• ISBN: 1467343935; 9781467343930
• ISSN: 2167-2121
• DOI: 10.1145/1957656.1957739

Personal robots have many things to learn and require a large quantity of data to learn them. Whether learning by demonstration, by trial and error, or collecting datasets for perception, robots will need to collect vast amounts of data without burdening the subjects. The parallels between gathering data for robot training and observing users during studies suggest the application of user study methodology as a basis for data collection methodology. Given the wide array of possible data, robotic platforms and algorithms, it is too early to set strict guidelines on collection practices. A clear set of guidelines, however, on how to report collection methodology and possible biases would benefit the community.