The Critterbot robot is designed to operate autonomously for long periods of time. Operating autonomously allows a robot to gather sensorimotor data in an unsupervised manner, without the need for human interaction. In this project, we study how data gathered during autonomous operation may be used to improve the Critterbot’s learning efficiency. The broader problem of transfer learning – reusing knowledge from one task to the next – has been mostly studied in small environments and domains where the available data is limited. A key aspect of this work is to develop new transfer learning algorithms that can handle months of robot data.