A bio-inspired control system and a VRML Simulator for an Autonomous Humanoid Arm. Michele Folgheraiter and Giuseppina Gini DEI Department,Politecnico di Milano, piazza Leonardo da Vinci 32, Italy Abstract. A hierarchical control architecture and a virtual model are under de- velopment on the basis of a biologically based analysis of manipulation. The control model has two levels: immediately over the effectors, the motoneurons give directly commands. Over it, the pattern generator and the inverse kinematic neural network transforms specifications of points to reach with the hand into joints values to actuate.The model of the virtual arm, integrated with the control model, helps to better understand the simulation results. 1 Introduction A Humanoid Robot, in order to mimic the human morphology and functionality, needs a complex kinematic structure. Since in the human body we have more than one hundred degrees of freedom, a hu- manoid robots needs in theory one hundred actuators at least, and a double number of sensors to sense position and force in each actuator. Moreover, it is extremely important in a humanoid robot to control the joint compliance in order to allow a safe collabo- ration with the human beings. In this situation classical control systems [1, 2], based on the accurate knowledge of the process dynamic model, are not so easy usable. This observation is more strong if the robot that we want to control is composed by light and high power to weight ratio actuators, like McKibben pneumatic artificial muscles or electroactive polymers fibers. This kind of devices tend to change their comportment when the number of working cycles increase, therefore a classical PID controller is not able to maintain the initial performance and an adaptable control system is needed. The control system strategy and the arm model,proposed in this paper, have been de- signed for our prototypes of artificial hands, namely Blackfingers and Whitefingers, illustrated in Figure1. The first project was born at Politecnico di Milano three years ago, while the second derived from the collaboration between our AIRLAB Laboratory and the Intelligent Robotics Laboratory of the Portland State University, started with the aim to conduct coordinated experiments. Both artificial hands were designed on the basis of anatomical studies of the human limb. Each hand is composed by five fingers with 4 mechanical degrees of freedom (DOF) each. In particular the first phalanx is provided by a spherical joint wraps by an elastic band that allow only two DOF. The joints between the first and the second phalanx and between the second and third phalanx are cylindrical, and they permit only the rotation respect the axial axis. In order to test our control system we have developed also a virtual model for the arm. We used this model for the two following purposes: