Assessing Humanoid Multimodal Grasping Towards Object Recognition

In this paper we present the outcome of a reflex-based humanoid grasping behaviour and assess its capacity to be used towards object recognition through acquiring multimodal information. Sensory data collected during the grasping activity of the iCub's hand is used to assess the relevance of proprioception and tactile information in building object representations that can be used to ultimately distinguish between objects. Results show that proprioception coupled with the proposed grasping patterns construction is most reliable as compared to tactile information. Hence, the work highlights the importance of improving the reliability of the tactile modality.