An approach to anthropomorphic robotics: guidelines and experiments

Abstract

If we could obtain robots which act as humans, does this imply that both have the same structure and are made of the same "substance"? The paper addresses this open question by implementing a basic robotic motor task using anthropomorphic hardware and an anthropomorphic computational approach. Two different algorithms implementing reinforcement learning both at a psychological and bio-mimetic level are presented and discussed. The performance achieved during the experimental trials suggests that similar performance can be achieved by the formal mathematical model of reinforcement learning and by the locally bio-mimetic architecture. This is very promising for the achievement of the adaptability, versatility and flexibility required of a humanoid, whilst still at acceptable levels of computational burden.