Teaching Robots to "Get It Right"

Abstract

We are interested in building and deploying service mobile robots to assist with arbitrary end-user tasks in everyday environments. In such open-world settings, how can we ensure that robots 1) are robust to environmental changes; 2) navigate the world in ways consistent with social and other unwritten norms; and 3) correctly complete the tasks expected of them? In this work, we survey these technical challenges and present several promising directions to address them. To "get it right", robots will have to reason about unexpected sources of failures in the real world and learn to overcome them; glean appropriate contextual information from perception to understand how to navigate in the world; and infer what correct task execution actually entails.