Humans have an outstanding ability to reason about space and the spatial behavior of other people. This ability facilitates both implicit and explicit interactions, e.g., navigation nearby other people as well as social conversations. As mobile robots become more capable in many domains, like health and entertainment, this kind of spatial reasoning becomes essential for them to integrate with and operate among people.

In this talk, I will present recent efforts to advance computational spatial reasoning in order to enable more effective social robot navigation and human-robot interactions in unstructured environments. Some of these efforts are focused on fundamental perception abilities for mobile robot operation, like traversability estimation and conversational group detection. Others focus on joint perception and action for robot navigation. Overall, these efforts demonstrate the potential of machine learning to help advance social robotics, and emphasize the importance of studying group human-robot interactions.

Several of the projects that will be presented in the talk are in collaboration with the JackRabbot group within the Stanford Vision & Learning Lab.