Experiments in synthetic psychology for tactile perception in robots: steps towards implementing humanoid robots

摘要

A robotic system which imitates the development process of stable grasping in infants is presented. The sensing devices of the system are tactile and visual sensors based on anthropomorphic design. The sensory data are processed and fused by the internal learning system based on artificial neural networks implementing the psychological aspects of the reinforcement learning paradigm. The design of the robotic system is modelled on the principles of synthetic psychology in order to obtain a human-like behavior, an important step towards "humanoids". Experiments show that the robot is able to find the best procedure to hold an object using the learned force with an average of 9 successful grasps over 10 trials.