Efficient algorithms for low-effort human teaching of robots

摘要: As autonomous robots begin to enter homes and workplaces, end-users of robots will gradually shift from trained robotists to non-experts with little knowledge of the underlying technology. To adapt to unstructured environments and diverse users, robots can be equipped with machine learning algorithms that allow humans to teach or guide them to acquire new skills. However, teaching is a complex task itself and robot-teaching is burdensome especially for naive end-users that are not familiar with the technology. Developing methods that help robots to learn in a manner natural to humans will allow deployed robots to efficiently learn new skills and improve performance without overburdening their human partners.