A biologically inspired robotic model for learning by imitation

Abstract

This paper proposes a biologically inspired model for motor skill imitation. The model is composed of modules that are high-level abstractions of the spinal cord, the primary and pre-motor cortex, the cerebellum, and the temporal cortex. Each module is modeled at a connectionist level. Primary motor behaviors, such as rhythmic movements of arm and legs for open-loop walking, are predefined in the spinal cord. Learning of new combination of movements is done by the DRAMA [8] neural architecture, which allows learning of time series and of spatio-temporal invariances in multimodal inputs. The model is implemented in a mechanical simulation of two humanoid avatars, the imitator and the imitatee 1 . We present three types of sequence learning: 1) repetitive patterns of arm and leg movements; 2) oscillatory movements of shoulder and elbows; 3) precise movements of the extremities for grasping and reaching. 1. INTRODUCTION A better understanding of the neurological substrate of learning ...