Visual Capabilities in an Interactive Autonomous Robot

Abstract

We build visually guided mobile robots that collaborate with us in dynamic unstructured environments. While engineering these embodied agents we discover what visual information is relevant for a variety of tasks. When we build systems that interact with the world, with humans, and with other agents, we rely upon all of the aspects of cognitive vision.

Our Robot Partners project at UBC focuses on the design and implementation of interacting autonomous mobile robots. The capabilities of the robots that permit collaboration in a variety of roles include mapping and localization, recognizing people, interpreting gestures and expression, naming scenes and interacting with people.

The roles the robots can play include José, the robot waiter, and HOMER, the Human Oriented Messenger Robot, both of which involve human-robot interaction. Our recent work aims to extend these roles and to learn useful categories of visual signals in the context of particular tasks.