The results of our previous research on the pedagogical use of Speaking Robots (SRs) revealed positive effects on motivating students to practice their oral skills in a stress-free environment. However, our findings indicated that the SR was sometimes unable to understand students' foreign accented speech. In this paper, we report the results of a study that investigated the ability of an SR to recognize and process non-native English speech from different levels of accentedness. The analysis is based on how the SR handled the participants' speech in terms of accuracy, the number and types of communication breakdowns observed, and how the participants behaved to solve the interaction problems that they experienced with the SR. Based on the study's surveys, interviews, and observations of users' interactions with the device, the results emphasize SRs' potential to recognize different types of accented L2 speech and their use as pedagogical tools. [For the complete volume, see ED578177.]