A behavior-based interactive learning approach for HRI: Teaching a robot new tricks

Many robots are pre-programmed for specific tasks/environments and perform poorly in real-world environments with diverse end-user demands. We propose a behaviorbased interactive learning approach that enables an end-user to “program” a robot for dynamic tasks. The end-user can teach the robot new primitive behaviors or improve pre-programmed ones, which make up hybrid higher level behaviors. The interface enables the end-user to learn about the robot’s capabilities and choose the right mixture of robot autonomy and human intelligence for the task at hand. We apply this approach to two application domains: searchers operating a UAV in Wilderness Search and Rescue and therapists preparing a robot for treating children with Autism. We show the interface is easy to use and enables a robot to perform variety of tasks with improved task performance and user satisfaction. We identify principles in HRI through user studies and evaluate how they affect the interactive learning process.