Recent developments in philosophy, linguistics, developmentalpsychology and artificial intelligence make itpossible to envision a developmental path for an artificialagent, grounded in activity-based sensorimotorrepresentations. This paper describes how Neo, an artificialagent, learns concepts by interacting with itssimulated environment. Relatively little prior structureis required to learn fairly accurate representationsof objects, activities, locations and other aspects ofNeo's...