Novel Tactile Descriptors and a Tactile Transfer Learning Technique for Active In-Hand Object Recognition via Texture Properties

Mohsen Kaboli , Gordon Cheng (1)

1 Xi’an Institute of Optics and Precision Mechanics

This paper proposes robust tactile descriptors and ,for the first time, a novel online tactile transfer learning strategy for discriminating objects through surface texture properties via a robotic hand and an artificial robotic skin. Using the proposed tactile descriptors the robotic hand can extract robust tactile information from generated vibro-tactile signals during in-hand object exploration. Tactile transfer learning algorithm enables the robotic system to autonomously select and then exploit the previously learned multiple texture models when classifying new objects with a few training samples or even one. The experimental outcomes demonstrate that employing the proposed methods and 10 prior texture models, the robotic hand could identify 12 objects through their surface textures properties with 97% and 100% recognition rate respectively with only one and ten training samples.

Mots clés: Tactile Transfer Learning; Tactile Object Recognition; Active Tactile Exploration; Tactile Descriptors; Tactile Signal Processing; Tactile Online Learning

Domaines: Robotique [cs.RO]; Intelligence artificielle [cs.AI]; Bio-informatique [q-bio.QM]; Informatique

Mohsen Kaboli, Gordon Cheng. Novel Tactile Descriptors and a Tactile Transfer Learning Technique for Active In-Hand Object Recognition via Texture Properties. IEE-RAS International Conference on Humanoid Robots-Workshop Tactile sensing for manipulation: new progress and challenges, IEEE, Nov 2016, Cancun, Mexico. ⟨hal-01385113⟩