Breazeal, C. & Scassellatie, B. (2002). Robots That Imitate Humans.

Social learning is an important feature that every robot should have. It is, however, extremely difficult to find the optimal way to implement social learning. One potential way for robots to learn, is to use imitation. The robots can learn by observing what other robots and/or humans are doing.

There are two main issues when it comes to social learning:

1. What should the robot imitate?

2. How should the robot imitate?

The question of what the robot should imitate is a difficult one, since it involves concepts like perception and attention. Even if the robot can perceive actions of the demonstrator, how does it know what aspects of the demonstrator’s behaviour it should attend to, and what it should ignore?

Secondly, there is the question of how the robot should imitate what it has just perceived. This is where the correspondence problem comes into play. The correspondence problem refers to the gap between perceiving what to imitate, and actually translating this perception into motor commands. As of right now, there is no obvious solution, but there are two different possible ways to achieve the proper movement: representing the perceived movement in motor-based or task-based terms.

In the end, the authors conclude that imitation-inspired mechanisms have played three dominant roles in robotics research:

1. Imitation can be an easy way to program a robot to perform novel actions.

2. Imitation can be a communicating mechanism (robot-human or robot-robot).

3. Imitation has been an effective tool for efficient motor learning in high-dimensional spaces.