Compliant Non-Prehensile Pushing Manipulation

Compliant Non-Prehensile Pushing Manipulation Francesco Cufino, Mario Selvaggio, Fabio Ruggiero PRISMA Lab, DIETI, University of Naples Federico II Via Claudio, 21, Naples, 80125, Italy [email protected] Fabio Amadio, Inria, Université de Lorraine, CNRS, LORIA, 5400, Nancy, France Abstract— In this work, we perform non-prehensile object pushing with a compliant robotic manipulation system to ensure safe operations in human-populated environments. We extend a state-of-the-art model-based controller to realize a desired pushing force while varying the contact point using an impedance-controlled robot. We then integrate an energy tank passivity-based control framework to modulate the velocity setpoint in a way that preserves passivity of the system. The proposed method has been rigorously tested in simulation and validated through real-world experiments. Keywords—Dexterous Manipulation; Non-Prehensile pushing; Compliance and Impedance Control; Physical HumanRobot Interaction I. INTRODUCTION SERVICE robots applications are becoming increasingly numerous and varied across multiple fields, supporting humans in executing tasks, often operating in unstructured and dynamic environments [1]. One example is represented by scenarios like hospital logistics (see Fig. 1), in which robots must be capable of physically interacting with healthcare personnel, to transport sanitary material, or with patients, to deliver them food. For the purpose of this work, all the scenarios in which robots physically manipulate objects in human-populated environments are particularly relevant. A. Motivation Introducing service robots in human-populated environments requires human-like dexterous manipulation abilities and compliant, and potentially safe, physical humanrobot interaction. In this context, non-prehensile manipulation enables a wide range of flexible and versatile actions for robots [2], [3]. This work focuses its attention on pushing manipulation primitive [2]. Several non-prehensile pushing techniques have been proposed in the past, although the possibility of deploying these in human-populated environments has remained largely unexplored. Our case study consists of a robot pushing a rack containing test tubes to deliver it to a human operator, thus facilitating their processing operations. Our aim is to investigate how to ensure safe physical interaction via control. B. Contributions The work presents the following contributions: (1) The work in [4] is extended to simultaneously realize a pushing force on an object and change the contact point using an impedance-controlled robot suitable for human-robot interaction scenarios; (2) The energy tank passivity framework is exploited for non-prehensile pushing, ensuring a limited exchange of energy within the sub-systems to prevent potential passivity violations when interacting with humans; (3) The devised framework is tested in both a physics-engine simulator and on two real robotic systems in a relevant hospital logistics task; (4) Open-source code for the developed framework is provided at the following link. Fig. 1. Picture showing the scenario enabled by our work: a robot pushes a rack of vials in a hospital environment while a human operator safely interacts with it thanks to our control framework. II. RELATED WORKS The first physics-based model for non-prehensile pushing was presented in [5], establishing a mapping from the center of pressure to the object’s angular velocity. Under quasi-static conditions, pushing forces relate to object velocities via the limit surface, a bounded convex set of friction loads [6]. Its ellipsoidal approximation [7] is widely used in planning and control. Recent methods for solving non-prehensile pushing employ Model Predictive Control (MPC), capturing the problem’s hybrid nature with complementarity constraints, later relaxed to avoid the use of mixed-integer programs [4]. When a dynamic interaction occurs among the robot, the environment and/or the human, energy into the system could be injected. To preserve a passive behaviour, the introduction of energy flow monitoring elements in the controller, also known as energy tanks, was proposed in [8]. We integrate such a mechanism for compliance set-point regulation, preserving passivity and enabling safe human interaction. III. CONTROL FRAMEWORK A. Compliant pushing We start from the control framework proposed in [4], keeping the planar-pushing model and optimization-based controller with complementarity constraints, and extend it to an impedance-controlled manipulator by adding a virtual spring model in the MPC and a passivity filter. Our control scheme is shown in Fig. 2. Using a virtual spring model (neglecting damping due to quasi-static conditions), we aim to realize both an arbitrary pushing force and a contact sliding velocity. Yet, since we control the planar position of the free end of the spring (2 variables), and the desired output involves 3 variables (2 force components and 1 sliding velocity), the system is underactuated. Thus, we cannot independently impose both force and sliding motion. Instead, we must derive a position/velocity reference that balances these objectives. This results in a coupled control of force and sliding velocity. The use of complementarity constraints with a slack variable allows smooth transitions between contact modes, enabling his trade-off during task execution [4]. Our strategy consists 2025 I-RIM Conference October 17-19, Rome, Italy ISBN: 9788894580570 10.5281/zenodo.17629802 171 in deriving a position/velocity reference that indirectly controls both the force and the sliding velocity, balancing between their respective optimal values. So, the formulated MPC takes as input the reference trajectory and the feedback of the actual state of the slider object and gives as output the planar position/velocity end-effector set-point which considers both the objectives. B. Passivity filter If an external interaction were to prevent adequate tracking of the object reference trajectory, the MPC would attempt to increase the pushing force to compensate for the growing tracking error. This is an unsafe behaviour which can be seen as an indefinite generation of internal energy. To prevent this, we make the MPC output pass through a passivity filter. This is realized introducing an energy tank whose goal is to recover the dissipated energy and use it to realize a less-conservative control action without violating the passivity of the system. IV. RESULTS A. Simulation The proposed framework has been first validated in simulation, where an interaction scenario is reproduced while an impedance-controlled robot performs the pushing task of an object along an eight-shaped trajectory. The robot we use is the 7-DOF KUKA LBR iiwa 7, commanded in torque and equipped with a steel stick with a spherical-shaped end. The simulation shows that the presented control scheme is able to maintain high tracking performance in the non-prehensile pushing task, despite the robot being impedance-controlled: in terms of absolute value, both the components of the object position tracking error do not exceed 1.5 × 10−2m, while the orientation one is below 10−2 rad. Furthermore, we demonstrate that the presence of the passivity filter ensures the robot to behave as expected, modulating the set point velocity (and so the pushing force) when potentially unbounded internal energy is generated leading to unsafe behavior. B. Experiments The experiments have been carried out on two different robotic systems: ABB Mobile Yumi, to test passive interaction scenarios, and KUKA LBR iiwa 7, to analyze trajectory tracking performance in presence of environmental uncertainties. The robots are equipped with cylindrical and spherical end-effector to push the object, respectively, and with in-hand cameras to perceive the object state. The experiments on mobile Yumi consist in pushing along a linear and a curvilinear trajectory, with an external interaction occurring during the task. The object tracking position error is of the order of 10−3m, and the orientation one is of the order of 10−2 rad, demonstrating acceptable performance. It is shown then that during the interaction, the tank energy decreases until it reaches the lower bound and the passivity filter turns off the control velocity action. After the human releases the object, the robot moves again, the tank recharges, and the task can be completed. The experiments on KUKA LBR iiwa 7 have been performed to assess performance and robustness varying linear object-table friction coefficient, object mass, object desired path, and desired trajectory mean velocity up to performance limits. The statistically significant performance worsening is observed either with higher rotational friction, with a lower mass, on curvilinear path, or with higher velocity. However, in summary, with a general mean positional error of 0.0165 m and mean rotational error of 0.1003 rad in all the perturbed cases, our method exhibits performance comparable to most recent pushing approaches (e.g., [9]) where RMS positional errors of 0.0052 m (mean x-axis and y-axis) and rotational one of 0.21 rad are reported. A video summarizing the results is available at this link. V. CONCLUSION Our extension of the state-of-the art optimal control problem for object pushing presented in [4] has allowed an impedance-controlled robot to perform pushing manipulation with good performance, realizing at the same time a pushing force while changing contact point. The robot can safely execute non-prehensile pushing tasks in a human-centered environment, taking into account every external interaction, either voluntary or not. Several future developments could be implemented, like removing the quasi-static assumptions, considering object shape variation, or introducing force feedback in the framework. REFERENCES [1] I. Lee, “Service robots: A systematic literature review,” Electronics, vol. 10, p. 2658, 2021. [2] F. Ruggiero, V. Lippiello, and B. Siciliano, “Nonprehensile dynamic manipulation: A survey,” IEEE Robot. Autom. Lett., vol. 3, no. 3, pp. 1711–1718, 2018. [3] M. Selvaggio, A. Garg, F. Ruggiero, G. Oriolo, and B. Siciliano, “Nonprehensile object transportation via model predictive non-sliding manipulation control,” IEEE Trans. Control Syst. Technol., vol. 31, no. 5, pp. 2231–2244, 2023. [4] J. Moura, T. Stouraitis, and S. Vijayakumar, “Non-prehensile planar manipulation via trajectory optimization with complementarity constraints,” in 2022 Int. Conf. Robot. Autom., pp. 970–976, 2022. [5] M. T. Mason, “Mechanics and planning of manipulator pushing operations,” Int. J. Robotics Res., vol. 5, no. 3, pp. 53–71, 1986. [6] S. Goyal, A. Ruina, and J. Papadopoulos, “Limit surface and moment function descriptions of planar sliding,” IEEE Int. Conf. Robot. Autom., pp. 794–799, 1989. [7] S. Lee and M. Cutkosky, “Fixture planning with friction,” J. Manuf. Sci. Eng., vol. 113, no. 3, pp. 320–327, 1991. [8] V. Duindam and S. Stramigioli, “Port-based asymptotic curve tracking for mechanical systems,” Eur. J. Control, vol. 10, no. 5, pp. 411–420, 2004. [9] S. Federico, M. Costanzo, M. De Simone, and C. Natale, “Nonlinear model predictive control for robotic pushing of planar objects with generic shape,” IEEE Robot. Autom. Lett., vol. 10, no. 3, pp. 3006– 3013, 2025 Fig. 2. Control scheme. 172