Abstract² The ability to grasp and manipulate objects can be regarded as a distinctive feature of humans in the animal world. The human hand is studied for different purposes, such as the control of anthropomorphic robotic hands. One of the fundamental features of the human hand is the thumb opposition, not always considered as active degree of freedom in the design of robotic hands. The purpose of this study is to investigate the role of thumb opposition during cyclic manipulation tasks through the interaction with different objects and a bio-inspired control architecture based on reinforcement learning. The control architecture has been implemented in simulated environment on two robotic hands with different thumb features, i.e. the iCub hand and the DLR/HIT Hand II, interacting with objects of different sizes and shapes. Furthermore, a systematic analysis of the working areas of the two hands during the manipulation of a sphere located at different positions has been carried out. The achieved results show that thumb opposition, characterizing human hands, plays a key function in fine manipulation and allows increasing the hand working area. I. INTRODUCTION he ability of finely manipulating and using objects as tools can be regarded as a distinctive feature of humans in the animal world. The complex structure of the human hand is at the basis of these manipulation capabilities [1]. The human hand is studied in robotics for different purposes, such as design of the new mechanical components and control. In particular, the analysis of the role of the kinematic structure of the human hand, and the role it plays in manipulation capabilities, is fundamental for the realization of multifingered anthropomorphic robotic hands [2]-[6]. The literature shows how human manipulation capabilities are heavily related to a peculiar feature of the human hand: thumb opposition [7] [8]. Thumb opposition This work was supported by the national project PRIN 2008 ‡23(1+$1’ - OPEN neuro-SURWKHVLF +$1’ SODWIRUP IRU FOLQLFDO WULDOV· &83 %- DQG ‡,7,1(5,6 · RQ WKH WHFKQRORJLFDO WUDQVIHU CUP: F87G10000130009. It was also supported by the EU funded project ‡,0-CLeVeR - Intrinsically Motivated Cumulative Learning Versatile 5RERWV· JUDQW DJUHHPHQW ,&7-IP-231722, FP7/2007-2013 ''Challenge 2 - Cognitive Systems, Interaction, Robotics''. Anna Lisa Ciancio, Loredana Zollo and Eugenio Guglielmelli are with Laboratory of Biomedical Robotics and Biomicrosystem, Università Campus Bio-Medico di Roma Via Álvaro del Portillo 21, I-00128 Roma, Italy (e-mail: {a.ciancio, l.zollo, e.guglielmelli}@unicampus.it). Daniele Caligiore and Gianluca Baldassare are with Laboratory of Computational Embodied Neuroscience, Istituto di Scienze e Tecnologie della Cognizione, Consiglio Nazionale delle Ricerche (LOCEN-ISTC-CNR) Via San Martino della Battaglia 44, I-00185 Roma, Italy (e-mail. {daniele.caligiore, gianluca.baldassarre}@istc.cnr.it). particularly attracts attention because of its fundamental role in hand functions [8] [9]. In this respect, clinical and biomechanical studies show that the thumb is responsible for 50% of hand functions [10]. The thumb opposition enables the human hand to carry out prehensile movements and hold objects firmly. It is also important for everyday life movements involving object grasping and manipulation, such as writing, turning a key, etc. The human thumb has been described by different kinematic models [11]-[13] with different characterization of the thumb joints. One of these models represents the thumb with 5 degrees of freedom: flexion/extension of the IP joint, considered as hinge joint, flexion/extension and adduction/abduction of the Metacarpophalangeal (MCP) joint, and flexion/extension and adduction/abduction of the Carpometacarpal (CMC) joint. Despite the literature on robotic hands presents a wide variety of configurations and kinematic models of the thumb, only a few of them have thumb opposition as an active DOF [4] [14]. For instance, in the Utah/MIT hand the thumb is mounted directly opposed to the other fingers in a fixed position [15], while the UB hand is characterized by an opposable thumb with 4 actuated degrees of mobility [16]. The purpose of this work is to investigate the role of thumb opposition in tasks of cyclic manipulation acquired through exploration and learning, based on the bio-inspired neural architecture proposed in [17]. We used this model to learn controlling two different anthropomorphic robotic hands engaged with manipulation of different objects: the iCub hand [3] and the DLR-HIT Hand II [5]. The main distinction between them concerns the thumb: the iCub hand has thumb opposition as an active DOF whereas the DLR/HIT Hand II has a fixed thumb opposition. The study was conducted using two 3D simulated robotic hands interacting with 3D simulated objects. The hands were controlled with the bio-inspired neural architecture proposed in [17]. It is based on two key elements: a hierarchical reinforcement learning neural network and Central Pattern Generators (CPGs). The neural network implemented the trial-and-error mechanisms guiding the system learning and the hierarchical model permitted to search parameters of different CPGs. The CPGs allowed producing rhythmic patterns to move the fingers during the manipulation tasks. This hierarchical bio-inspired control was applied to the two simulated robotic hands interacting with 9 different objects of different shapes and dimensions. In particular, the task required the hands to rotate each object around a fixed axis T The Role of Thumb Opposition in Cyclic Manipulation: A Study with Two Different Robotic Hands A.L. Ciancio, L. Zollo, Member, IEEE, G. Baldassarre, D. Caligiore, E. Guglielmelli, Senior Member, IEEE