Active Motor Babbling for SensoryMotor （活跃的电动机为SensoryMotor胡说）.pdfVIP

Active Motor Babbling for Sensory-Motor Learning∗ Ryo Saegusa, Giorgio Metta, and Giulio Sandini Sophie Sakka Robotics, Brain and Cognitive Sciences Department Laboratory of Solids Mechanics Italian Institute of Technology University of Poitiers Via Morego 30, 16163 Genoa, Italy BP30179-86962 Futuroscope Chasseneuil Cedex, France ryos@ieee.org, pasa@liralab.it, giulio.sandini@iit.it sophie.sakka@lms.univ-poitiers.fr Abstract—For a complex autonomous robotic system such a modular control approach [3], which couples a forward as a humanoid robot, the motor-babbling based sensory-motor model (state predictor) and an inverse model (controller). learning is considered effective to develop an internal model The forward model predicts the next state from a current of the self-body and the environment autonomously. In this paper we propose a methodology of sensory-motor learning state and a motor command (an efference copy), while and its evaluation towards active learning. The proposed model the inverse model generates a motor command from the is characterized by a function called confidence, which works current state and the predicted state. Even if a proper motor as a memory of reliability for state prediction and control. command is unknown, the feedback error learning procedure The confidence for the state can be a good measure to bias (FEL) provides a suitable approximation [4]. The prediction the next exploration strategy of data sampling, such as to direct its state to the unreliable domain. We consider the error contributes to gate lear