Learning to Solve Insertion Tasks Using Exploratory Behaviors

The shape-sorter puzzle is a quintessential toy that children play with to learn how to fit objects in holes. Psychology research shows that human children learn object fitting during an early developmental period. We implemented similar developmental learning in our robot using random exploratory behaviors and proprioceptive feedback with the shape sorter puzzle. Analysis shows that successful fits can be detected from proprioceptive feedback streams. Ultimately, we hope to show that this developmental approach can enable systems to be more robust and capable of accurately detecting and eventually predicting successful insertions in new, more complicated peg-in-hole situations.