The Agent-based perspective

Imitation is a powerful learning mechanism and a general agent-based approach must be used in order to identify the most interesting and significant problems, rather than the prominent ad hoc approaches in imitation robotics research so far. The traditional approach concentrates in finding an appropriate mechanism for imitation and developing a robot control architecture that identifies salient features in the movements of an (often visually observed) model, and maps them appropriately (via a built-in and usually static method) to motor outputs of the imitator (Kuniyoshi et al., 1994, 1990). Model and imitator are usually not interacting with each other, neither do they share and perceive a common context. Effectively this kind of approach limits itself to answering the question of how to imitate for a particular robotic system and its particular imitation task. This has led to many diverse approaches to robot controllers for imitative learning that are difficult to generalize across different contexts and to different robot platforms. In contrast to the above, the agent-based approach for imitation considers the behaviour of an autonomous agent in relation to its environment, including other autonomous agents. The mechanisms underlying imitation are not divorced from the behaviour-in-context, including the social and non-social environments, motivations, relationships among the agents, the agents individual and learning history etc. (Dautenhahn and Nehaniv, 2002).

Such a perspective helps unfold the full potential of research on imitation and helps in identifying challenging and important research issues.