Attention focus in curious, reconfigurable robots

The study of computational models of motivation - such as curiosity and interest - has opened the way for new types of artificial agents that can self-select their own goals and focus of attention. In robotic systems, curious agents that can select their own goals have a range of potential applications. These include support for tool use, fault tolerance and robot reconfigurability. This paper considers the design of curious, reconfigurable robots that can explore new behaviour in response to changes in their physical structure. A computational model of curiosity is combined with neural-network reinforcement learning to create curious, reconfigurable robots on the Lego Mindstorms NXT platform. Results show that the curious robot can shift its attention focus to learn new behaviours in response to changes in its structure.