Synthesis of Robot Hand Skills Powered by Crowdsourced Learning
Document Type
Conference Proceeding
Publication Date
5-24-2019
Abstract
Crowdsourcing has shown great potentials in artificial intelligence. Continuous learning from a large group of mentors breaks the limit of learning from one or a few mentors in individual cases, and has achieved success in image recognition, translation and many other cyber applications. We bring the power of crowdsourcing to robot physical intelligence and introduce a learning method that allows robots to synthesize new physical skills using knowledge acquired from crowd-sourced human mentors. In addition, we provide a solution to sustainably manage a continuously growing massive knowledge library. The method is validated using a virtual reality interface and a simulated test of robot in-hand manipulation. The work has the potential of robotizing many demanding tasks that are currently hard to automate due to the demanding requirement of hand skills. The effectiveness of crowdsourced learning is evaluated by studying the success rate of new skill synthesis and the performance of the synthesized skills.
Identifier
85067116690 (Scopus)
ISBN
[9781538669594]
Publication Title
Proceedings 2019 IEEE International Conference on Mechatronics Icm 2019
External Full Text Location
https://doi.org/10.1109/ICMECH.2019.8722953
First Page
211
Last Page
216
Recommended Citation
Zhao, Leidi; Lawhorn, Raheem; Wang, Cong; Lu, Lu; and Ouyang, Bo, "Synthesis of Robot Hand Skills Powered by Crowdsourced Learning" (2019). Faculty Publications. 7586.
https://digitalcommons.njit.edu/fac_pubs/7586
