The Learning With Kids Project: Retrospective and Status Report Victor Ng-Thow-Hing Honda Research Institute USA, Inc. Mountain View, CA, USA vng@honda-ri.com Ravi Kiran Sarvadevabhatla Honda Research Institute USA, Inc. Mountain View, CA, USA rsarvadevabhatla@hra.com Sandra Okita Columbia University New York, NY, USA okita@tc.columbia.edu Fig. 1. Robot helping a child learn to set a table. Abstract— The Learning with Kids (LwK) project is an ongoing collaboration with the long-term goal of building a humanoid robot that is capable of being a peer-based learning partner for children. Experiments for the first three years have recently been completed, and complete analysis of all data is currently in progress. We review the studies we have performed so far, the tools we developed to conduct these studies, and the adjustments we made as we learned along the way. I. I NTRODUCTION The Learning with Kids Project (LwK) consists of a series of human-robot interaction studies spanning a three year period from 2008-2010. Complete analysis of all data is in progress. The overall goals of this project were to develop methods to improve the engagement between the Honda humanoid robot and people, especially children (Figure 1), and to explore the use of humanoid robots as peer-based learning partners for educational applications. A. Background History Prior to this project, the group at the Honda Research In- stitute USA (HRI-US) had worked on developing interactive scenarios for human-robot interaction, such as the Memory Game project [4]. In that work, the focus was on solving technical integration issues and performance requirements necessary for successful turn-based interaction. We created communication middleware tightly integrated with a black- board architecture to enable efficient sharing of information between modules necessary for providing robot multi-modal perception and behavior [6]. For successful demonstration, our priority was creating robust perceptual algorithms for card and table detection (Figure 2). Consequently, constraints were placed on the robot’s behavior to produce stable per- ception such as keeping the robot’s head cameras focused only on the table during the entire game. Since interaction was turn-based, the robot would only interact with the player when it was the player’s turn and exclusively look at the cards and table during its own turn. From a social perspective, the robot appeared unfriendly and narrow-minded in its singular focus of playing the game. After this project, we Fig. 2. Memory Game desired to refine our interactive models to better engage human participants. In our trials with children playing the memory game, they grew bored quite quickly. Furthermore, prior to initiating the LwK project, the team at HRI-US had no experience with systematic evaluation studies between people and robots beyond testing within members of our research facility. B. Social Science Partnership To complement our technical perspective, we chose a partner who had experience running studies with children, yet who was also comfortable working with technology. Dr. Sandra Okita from Columbia University was an ideal match as her thesis explored the relationships between people and technology in various educational settings. In particular, she