CoZ: A crowd-powered system for social robotics

Abstract

We present Crowd of Oz (CoZ) — a crowd-powered system to enable a real-time dialogue between Softbank's Pepper robot and a user through a synchronous group of workers. We achieve this by: (1) providing media-rich interfaces and the underlying architecture to assist workers to contextualize the conversation; and (2) proposing and implementing the Pavilion algorithm, which ensures that the required number of workers are engaged. We have demonstrated in a lab setting that through CoZ, we can leverage crowdsourcing to drive the Pepper robot in a conversation on two topics: managing study-related stress and planning a holiday. Our preliminary results indicate that the crowd's input is both meaningful and fast enough to hold a conversation.