Friction Identification in a Pneumatic Gripper Rocco A. Romeo 1 , Marco Maggiali 1 , Daniele Pucci and Luca Fiorio 1 Abstract— Mechanical systems are typically composed of a number of contacting surfaces that move against each other. Such surfaces are subject to friction forces. These dissipate part of the actuation energy and cause an undesired effect on the overall system functioning. Therefore, a suitable model of friction is needed to elide its action. The choice of such a model is not always straightforward, as it is influenced by the system properties and dynamics. In this paper, we show the identification of different friction models and evaluate their prediction capability on an experimental dataset. Despite being state-of-the-art models, some modifications were introduced to improve their performance. A pneumatic gripper was used to collect the data for the models evaluation. Two experimental setups were mounted to execute the experiments: information from two pressure sensors, a load cell and a position sensor was employed for the identification. During the experiments, the gripper was actuated at different constant velocities. Results indicate that all the identified models offer a proper prediction of the real friction force. I. INTRODUCTION Friction is a fundamental quantity to be accounted for in mechanical engineering. It usually degrades the behavior of physical systems and, depending on the complexity of such systems, might complicate the design of control algorithms. Despite its effect is mitigable by means of lubricant films with a certain thickness, a proper model of friction is always required to effectively compensate for such an effect. Nonetheless, friction is quite difficult to model; this is known since several decades [1]. A number of models, characterized by varying complexity, were proposed so far. Among the classical ones, there is the Coulomb model, which dates back to more than two centuries ago. According to this model, when one surface slides over another one, the friction force F fd is proportional to the applied normal force F n through a constant, namely kinetic coefficient of friction μ d . When there is no sliding, the friction force can be as high as a value F fs = μ s F n , with μ s >μ d . This phenomenon is known as stiction. In all cases, the sign of the sliding velocity has to be involved in the friction force computation. When the system exhibits a certain degree of viscosity, i.e. friction force grows along with velocity ˙ x, a further term can be added in the form μ v ˙ x, being μ v the viscous coefficient. Thanks to its simplicity, the Coulomb model was largely employed in physics and engineering, and its usage is still quite common [2]. In general, simple models such as the Coulomb’s one are easier to implement but they do not cap- ture all the frictions phenomena, such as stick-slip. Moreover, when there is no movement between two contacting surfaces, the Coulomb’s model cannot predict friction. 1 iCub Tech, Istituto Italiano di Tecnologia, Via San Quirico 19D, 16163 Genoa, Italy. Email: firstname.lastname@iit.it To achieve higher detail, effects such as the Stribeck one are to be considered. A similar effect takes into account the aforementioned stick-slip [3], which occurs at low velocities and causes a non-linear drop in the friction force. Typically, the expression of the Stribeck phenomenon features an exponential function. Despite its greater completeness, even the Stribeck model does not resolve the discontinuity at null velocity introduced by the Coulomb model. In this paper, we intend to identify one or more suitable friction models for a real mechanical system, i.e. a pneumatic gripper. Even though there exist more complex compensation techniques (e.g. [4], in this initial study we concentrate on model-based approaches. Four models will be first identified and then evaluated, namely: Coulomb model with viscous friction (CV), Karnopp model [5], Threlfall model [6], and the Coulomb model with viscous friction and Stribeck effect (CVS). The models were selected based on their variegated characteristics: most probably, the highest degree of detail is provided by the CVS model which combines the Coulomb and viscous friction, along with the non-linearity typical of stick-slip. Some modifications were introduced to all the four models in order to remove the discontinuity at null velocity and/or to achieve superior performance. Notice that acronyms are given only for long names, i.e. only for CV and CVS models. It is also worth mentioning that all the selected models were static. i.e. they cannot work when ˙ x is non-constant. Dynamic models such as LuGre [1] will be investigated in successive studies. For all the models, the identification experiments were therefore conducted at constant velocity, on two different setups involving the gripper and two pressure regulators actu- ating it. Pneumatic grippers are still the most used in robotics [7], though friction was much more rarely investigated in these systems rather than e.g. in pneumatic cylinders [8]. Therefore, the aim of this study is to analyze friction forces in an off-the-shelf pneumatic gripper, providing insight on how such forces influence its functioning. Moreover, the understanding of friction would be of great help in the design of control strategies for pneumatic grippers, which still lack of reliable closed-loop force regulation [7]. In our previous works [7]-[9], we showed two force-control architectures that resorted to optimization algorithms [7], Kalman filters and state observers [9]. Nonetheless, the performance of such architectures was somehow limited by the lack of a friction model. The rest of the paper is structured as follows: Section II shows the pneumatic gripper adopted and the experimental setups, whereas Section III presents the friction models, which are all static models. Section IV illustrates the identi- 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) October 25-29, 2020, Las Vegas, NV, USA (Virtual) 978-1-7281-6211-9/20/$31.00 ©2020 IEEE 9948