Performance Analysis of a Mass-Spring-Damper Deformable Linear Object Model in Robotic Simulation Frameworks

Andrea Govoni 1, Nadia Zubair 1, Simone Soprani 1, Gianluca Palli 1 · 2025

Read Performance Analysis of a Mass-Spring-Damper Deformable Linear Object Model in Robotic Simulation Frameworks by Andrea Govoni 1, Nadia Zubair 1, Simone Soprani 1, Gianluca Palli 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-13659.txt.

Introduction

Andrea Govoni and three co-authors test how a spring-based model mimics bendy objects in robot simulators. Their 2025 study covers cables, wires, and strings, known as deformable linear objects. Such items are hard to simulate because they flex and twist freely. Good simulation matters for training robots to handle wires and assembly tasks.

The model treats the object as masses linked by springs and dampers. The team varied its settings and watched how the object's motion changed. Two simulators, Isaac Sim and Gazebo, hosted the tests. The goal was to gather force data with randomized settings for later robot training. The study checks whether such randomization works in these simulation setups.

Catalog record

CREATORAndrea Govoni 1, Nadia Zubair 1, Simone Soprani 1, Gianluca Palli 1
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE12 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-13659

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Performance Analysis of a Mass-Spring-Damper Deformable Linear Object Model in Robotic Simulation Frameworks Andrea Govoni 1, Nadia Zubair 1, Simone Soprani 1, Gianluca Palli 1 Abstract The modelling of Deformable Linear Objects (DLOs) such as cables, wires, and strings presents significant challenges due to their flexible and deformable nature. In robotics, accurately simulating the dynamic behavior of DLOs is essential for automating tasks like wire handling and assembly. The presented study is a preliminary analysis aimed at force data collection through domain randomization (DR) for training a robot in simulation, using a Mass-Spring-Damper (MSD) system as the reference model. The study aims to assess the impact of model parameter variations on DLO dynamics, using Isaac Sim and Gazebo to validate the applicability of DR technique in these scenarios. Code available at https://github.com/GovoUnibo/cable _ \_ model/settings † † tocauthor: Andrea Govoni, Nadia Zubair, Simone Soprani, Gianluca Palli † † institute: 1 Alma Mater Studiorum - Università di Bologna, Bologna 40126, IT, , , , 1 Introduction (a) Isaac Sim Simulation Setup (b) Gazebo Simulation Setup Figure 1 : wire setups of Isaac Sim and Gazebo Nowadays there is an increasing demand for automation in industries that handle DLO, such as automotive wire harness installation wireharness , cabling, and manipulation of flexible materials in manufacturing yang2022learning . Manipulating DLOs presents a unique challenge due to their inherent unpredictability under varying forces and constraints, which complicates robotic control lv2022dynamic . In these contexts, a thorough understanding of DLO behavior is essential, offering significant benefits, especially when simulation is used to optimize manipulation strategies. Several different DLO model approaches are available, each with its own advantages and challenges DBLP:journals/corr/abs-2108-08935 , lv2020review . For instance, the multibody approach rigid and finite element methods KAUFMANN2009153 offer detailed simulations but come with high computational costs ac2024 . Other models, such as those based on dynamic splines splines , article , provide alternative methods but without a physical approach. While these models provide valuable insights into the behavior of DLOs, their effectiveness and applicability can vary depending on the intended use. In contemporary research, cable models are increasingly employed in simulations to rapidly generate datasets for training purposes. Our goal is to utilize the DLO models described in lv2017physically , which discretize the DLO into point masses connected by dampers and springs. This MSD approach enables the collection of interaction force data alongside the cable’s position during manipulation with a dual robotic manipulator. These data serve as a foundation for refining control strategies, such as reinforcement learning (RL) SassanoRL , to enhance bimanual manipulation tasks. Achieving realistic DLO behavior through modeling is challenging. To address this, we plan to apply Domain Randomization (DR) in our simulations to improve the adaptability and performance of our control strategies in real-world scenarios DomainRandomization . DR involves training the model with a wide range of variations in simulation parameters. However, before implementing this approach, it is crucial to examine how these variations impact the model’s performance, particularly by identifying conditions that could lead to diminished effectiveness. Consequently, this work focuses on varying the parameters of the DLO model, with a specific emphasis on studying performance changes in robotic simulators. The paper is organized as follows: in Sec. 2 we describe the cable model with an emphasis on the parameters that influence its dynamics. Sec. 3 presents the results, iIllustrating how parameter variations impact the chosen Deformable Linear Object (DLO) model within the specific use case, presenting result validated using robotic simulators such as Isaac Sim and Gazebo. 2 Simulation Setup Parametrs Figure 2 : Cable discretized in n masses ( m_{i} ) connected by n-1 Segments ( l_{i} ) As shown in Fig. 1 , the DLO in simulation is represented as a discrete series of masses, each connected to adjacent masses through virtual springs (see Fig. 2 ). The behavior of mass i is described by Newton’s second law in Eq. 1 , which accounts for the elastic forces acting on it. m_{i}\frac{\partial^{2}x_{i}}{\partial t^{2}}+k_{d}\frac{\partial x_{i}}{\partial t}=-\frac{\partial E_{i}}{\partial x_{i}}+F^{e}_{i} (1) m_{i} represents the mass of the point, x_{i} is its position, k_{d} is the damping coefficient, E_{i} represents the total energy stored in the springs, and F^{e}_{i} are external forces such as gravity or interaction forces from the environment. In this framework, we will focus on linear springs and bending springs indicated respectively as l_{i} link in Fig. 2 and spirals in Fig. 2 . Twisting springs are excluded from the simulation, since they are unnecessary for our DLO manipulation purposes. Further details can be found in lv2017physically . 2.1 Linear and Bending Springs in the Simulation The energy stored in a linear spring ( E_{i}^{s} ) between two points is given by: \displaystyle E_{i}^{s} \displaystyle=\frac{1}{2}k_{s}(l_{i}-l_{0i})^{2} (2) \displaystyle F_{i}^{s} \displaystyle=-k_{s}(l_{i}-l_{0i})\mathbf{u} where, l_{i} is the current length between the mass points, l_{0} is the initial, resting length of the segment, and \mathbf{u} is a unit vector in the direction of the spring’s force. The stiffness k_{s} is calculated as: k_{s}=\frac{EA}{l_{0i}} (3) where, E is Young’s modulus and A is the cross-sectional area of the wire. The force F_{s} from the spring is then used to update the positions of the connected mass points at each time step. For bending, the energy associated with bending is given by: \displaystyle E_{i}^{b} \displaystyle=\frac{1}{2}k_{b}\beta_{i}^{2} (4) \displaystyle F_{i}^{b} \displaystyle=-\left(\frac{\partial E^{b}_{i-1}}{\partial x_{i}}+\frac{\partial E^{b}_{i}}{\partial x_{i}}+\frac{\partial E^{b}_{i+1}}{\partial x_{i}}\right) where \beta_{i} is the angle between segments, calculated as: \beta_{i}=\arctan\left(\frac{|\mathbf{l}_{i+1}\times\mathbf{l}_{i}|}{\mathbf{l}_{i+1}\cdot\mathbf{l}_{i}}\right) Here, \mathbf{l}_{i} is the vector representing the current segment between two mass points, and \mathbf{l}_{i-1} and \mathbf{l}_{i+1} represent adjacent segments. The bending stiffness k_{b} is computed as: k_{b}=\frac{EI}{l_{0i}} (5) where, I is the moment of inertia of the wire cross-section. These formulas are adapted for real-time force calculations among masses to update each simulation step. The computed forces are input for the simulator, which then uses integration, to update the DLO model’s velocities and positions. 2.2 Damping and Integrator Step Problem Choosing an appropriate damping coefficient ( k_{d} ) is challenging, as it must balance stability and responsiveness; a low k_{d} can result in excessive oscillations, while a high k_{d} may cause the system to become overdamped or even unstable. Although using variable time steps would improve stability and accuracy in complex numerical simulations, simulators like Gazebo and Isaac Sim often rely on fixed step sizes, making it necessary to choose corresponding k_{d} values. To ensure system stability with a fixed time step, we apply the following helping condition found in StabilityMSD : k_{d}\leq\frac{|vm\Delta t+F|}{|v|} (6) where v is the velocity, \Delta t is the fixed for time step for integration to update the simulation, and F is the internal forces. 3 Experiments & Conclusion Table 1 : Parameters used in the experiments No. E (MPa) m_{i} discretization \Delta t Sampling Period (s) Stability Sec To Instability 1 12.6 i=10 0.000005 Stable \inf 2 526.0 i=10 0.000005 Stable \inf 2 1002.0 i=6 0.000005 Stable \inf 3 1002.6 i=10 0.000005 Unstable 0.4 4 1002.6 i=10 0.0000001 Stable \inf Figure 3 : Analysis of the motion along the z-axis of the masses 2,4, moving an extremity of 5 mm, using E=12.6MPa, mass discretization = 6 and, 0.000005 Sampling period Figure 4 : Anlayzing masses 2,4,8, using E=1002.6MPa, mass discretization = 10 and, 0.00005 Sampling period

Read the complete text