Dexterous Hands Learn To Re-Use The Past Experience To Discriminate In-Hand Objects From The Surface Textures

This paper focuses on the problem of in-hand object recognition with anthropomorphic robotic hands via artificial robotic skin. Using our proposed batch tactile transfer learning technique, the robotic hands can re-use their prior tactile knowledge to discriminate in-hand objects while using few number of training samples. The experimental results show that the Shadow Hand could manage to classify in-hand objects with 100% recognition accuracy rate while using few training samples plus its past tactile experience.