From Perception-Action Loops to Imitation Processes: A Bottom-Up Approach of Learning by Imitation

Neural learning of embodied interaction dynamics

“…This way, we may be able to achieve a unity of embodied dynamics and various imitation abilities (Gaussier et al, 1994), not in a separate stages or components, but as intertwining dynamical processes. And by imitation, the agent can participate in rich interagent interaction, which will be essential for true achievement of communication, cooperation, and mutual understanding.…”

Learning behavior fusion from demonstration

“…This results in a general-purpose policy, which would allow the robot to perform the demonstrated task in any new environments, from any initial positions. We do not attempt to reproduce exact trajectories (such as in [28]), but rather learn the underlying policy for executing the task. In making arbitration decisions, sequential methods represent all possible coordinations as discrete set of possibilities along each axis of fusion space.…”

“…Successful approaches to second person LFD have demonstrated learning of reactive policies [27], [12], trajectories [28], words [29], and sequential task descriptions [30]. Second person approaches, where the teacher and learner act from different perspectives, often employ a teacher following strategy, where the robot learner follows a human or a robot teacher.…”

Robota: Clever toy and educational tool

“…The rational is that the robot's ability to imitate would offer a natural means of programming the robot, that would require none or minimal programming competence on the user's part. Robot learning by imitation, also known as robot programming through demonstration, finds its use for a large number of tasks, such as object manipulation [13,16], learning body motions [2,8,15], navigation [5,10] and learning communication skills [4].…”