Generation of rhythmic and voluntary patterns of mastication using Matsuoka oscillator for a humanoid chewing robot W.L. Xu a, * , F. Clara Fang b , J. Bronlund a , J. Potgieter a a School of Engineering and Technology, Massey University, Albany, Northshore, Auckland, New Zealand b College of Engineering, Technology and Architecture, University of Hartford, Connecticut, USA article info Article history: Received 8 October 2007 Accepted 19 August 2008 Keywords: Neural oscillator Chewing robot Robotic jaw Mastication Muscle activity CPG abstract We intend to apply Matsuoka neural oscillator into humanoid chewing robots to generate rhythmic actu- ation of central pattern generator (CPG) and adapt it for voluntary actuation due to sensory feedback. In this paper a single Matsuoka oscillator of two neurons is used for two phase-locked muscles (e.g. masse- ter and digastric muscles) or for a single robotic joint. To help design and tune the oscillator we have developed three graphical user interfaces (GUI) with aid of which the simulation, parameters’ influence and adaptation of the oscillator can be analysed and for specific pattern of muscle activities the oscillator can be selected. Discussions are made in relation to the experimentally confirmed EMG (electromyogra- phy) of muscle activities for various foods. A case study involving a jaw, driven by a couple of opening and closing muscles that are commanded by motoneurons is presented. The force of the muscles is described in nonlinear Hill model while the motoneuron for muscle activities is modelled in the oscillator. Simula- tions are performed to show the oscillator’s ability in generating and adapting its rhythmic outputs with respect to the chewing without food (i.e, EMG only for rhythmic muscle activities), with foods (i.e., EMG for rhythmic and additional muscle activities) and with crushable foods (to see how quickly the oscillator to reduce its force commands in order not to damage the teeth). Our work is also meaningful for brain- based control of assistive or rehabilitative devices and EMG-driven neuromusculoskeletal models. Ó 2008 Elsevier Ltd. All rights reserved. 1. Introduction Human mastication is continually modified throughout the chewing sequence in response to the food dynamics; and mastica- tion patterns vary between subjects and between food textures. This interactive relationship has been utilized to evaluate textural properties of foods by measurement of masticatory psychology [1,2]. The challenge in evaluating foods this way is that full physi- ologies of mastication can hardly be measured completely and the measurement interferes natural chewing to certain extent. As re- sult, the food texture can only be evaluated qualitatively and var- ious hypotheses could not be tested on human subjects easily. To this end, a chewing robot solution has been proposed [3,4]. The idea is that while being chewed by a robot, food properties and texture change during chewing are evaluated by robotic states in actuation, chewing force, and/or jaw movements. Although it could be used to chew foods, WJ (Waseda Jaw) robot series was developed to work especially with a WY series dental training ro- bot [5,6]. They did not take into account of the biological aspects of the human masticatory system; and hence the robotic states can not be used for purpose of foods evaluation. The chewing robot [7,8] was based on biomechanical findings about the jaw structure and muscles of mastication, and developed especially for foods evaluation. The above robots may be commanded to chew foods by following recorded masticatory movements and chewing forces, but not mimicking the way human does. Human chewing of foods is performed by the movement of the jaw that is actuated by the muscles of mastication. Alpha-moto- neurons (or alpha-MN) innervate a muscle by recruiting a number of motor units and firing them at various frequencies [9,10]. Elec- tromyography (EMG) measurements have confirmed that a small amount of muscle activities is required for the free rhythmic move- ments of the jaw, which is produced by central pattern generator (CPG); and additional voluntary muscle activities is generated in alpha-MN if the closing movement is resisted by foods [11,12] and the harder the food, the larger the muscle activities is required [13,14]. In principal, the muscle of mastication involves anticipa- tory (or feedforward) activities for pre-programmed movement depending on individual chewing expectation, rhythmic activities generated by CPG that is dictated by individual physiology, and voluntary (sensory feedback) activities for overcoming food resistance. To make a chewing robot chew human way, the above mentioned rhythmic, anticipatory, voluntary patterns of muscle activities should be implemented. 0957-4158/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.mechatronics.2008.08.003 * Corresponding author. E-mail address: W.L.Xu@Massey.ac.nz (W.L. Xu). Mechatronics 19 (2009) 205–217 Contents lists available at ScienceDirect Mechatronics journal homepage: www.elsevier.com/locate/mechatronics