Toward Learning to Write by Identifying Writable Surfaces

Writing is a powerful skill that humans use on a daily basis to convey and capture information. J.J. Gibson pointed out that writing is a special form of tool use, which requires a special tool that has the ability to leave a trace on a surface: “A hand-held tool of enormous importance is one that, when applied to a surface, leaves traces and this affords tracemaking. The tool may be a stylus, brush, crayon, pen, or pencil, but if it marks the surface it can be used to depict and to write, to represent scenes and to specify words” [1, p.134]. Learning to identify objects that could potentially leave traces on novel surfaces is an important skill that humans learn to master. For example, a pencil can leave a trace on paper but not on a white board. The human-assisting robots of the near future would be more useful if they knew how to use writing instruments and tools. This investigation asks the question: How can a robot learn to identify object-surface pairs that are useful for making traces? This is an initial step toward identifying a developmental sequence that a robot could use in order to learn how to write.