There are several areas where we think the Geo can significantly improve in, and
which we believe to be interesting research topics.

State Transparency: One of the biggest hindrances to people’s ability to teach
the Geo is their difficulty assessing which state the Geo is in. It would be
nice if this state information can be conveyed in a natural, social manner.

Natural Learning Behavior: We observed through this experiment that the
programmatic method of learning we utilized did not fit very well within the
context of a “natural” social interaction between the user and the robot. Users
complained that the learning behavior of the robot was hard to understand and
that it was easy to confuse the engagement actions (when engagement was turned
on) with the dance step actions they were supposed to teach.

Modeling: We observed that many users wanted to “model teach” the Geo, which
consists of manually shaping the Geo’s joints in a specific way and asking the
robot to “remember how to do that” in a way that is similar to key-framing. We
believe that this method of teaching might be something that is very natural for
humans, and thus might constitute an interesting area for future exploration.