IsaacIPC: Coupling High-Fidelity Simulation and Realistic Rendering for Contact-Rich Robotic Systems CCS: Computing methodologies Physical simulation CCS: Computing methodologies Robotics
Qixin Liang, Zhongqing Han · 2026
Read IsaacIPC: Coupling High-Fidelity Simulation and Realistic Rendering for Contact-Rich Robotic Systems CCS: Computing methodologies Physical simulation CCS: Computing methodologies Robotics by Qixin Liang, Zhongqing Han (2026) 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-2605-24339.txt.
Introduction
Liang and Han present IsaacIPC, a robotic sim tool coupling GPU contact physics with IsaacSim and IsaacLab. IsaacIPC maps deformation between sim and visual meshes, giving live lifelike visuals for data and policy tests. For tactile sensing, the mortar contact rule sets a barrier over contact samples on touch surfaces, resolving pressure patterns better. The authors test the rule on contact benchmarks and show IsaacIPC on rigid-soft sims including a four-legged robot, a dexterous hand, and a gripper.
Catalog record
| CREATOR | Qixin Liang, Zhongqing Han |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29356 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.24339
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
IsaacIPC: Coupling High-Fidelity Simulation and Realistic Rendering for Contact-Rich Robotic Systems CCS: Computing methodologies Physical simulation CCS: Computing methodologies Robotics Qixin Liang, Zhongqing Han Abstract . We present IsaacIPC, a robotic simulation framework that couples GPU accelerated incremental potential contact (IPC) with Isaac Sim / Isaac Lab. IsaacIPC maps simulated deformation between simulation and visual meshes, enabling real-time realistic rendering with applications to data collection and policy evaluation. For tactile sensing, we introduce the geometric mortar contact potential (GMCP), which defines a barrier potential over contact samples on tactile surfaces to better resolve contact-pressure distributions. We evaluate GMCP on contact benchmarks and demonstrate IsaacIPC on rigid–deformable robotic simulations including a quadruped robot, a dexterous hand, and a universal manipulation interface (UMI) gripper. Figure 1. IsaacIPC qualitative demonstrations across rigid–deformable robotic systems: a soft Universal Manipulation Interface (UMI) gripper, a dexterous hand, and a quadruped robot. Three qualitative IsaacIPC demonstrations showing a universal manipulation interface (UMI) gripper, a dexterous hand, and a quadruped robot. 1. Introduction High-fidelity robotic simulation frameworks are increasingly important for embodied intelligence: they enable scalable policy training, systematic evaluation, and contact-rich interactions that are expensive to collect in the real world ( 12 ; 27 ) . Contact simulation is central to these systems because manipulation, locomotion, and tactile sensing all rely heavily on physical interaction with the environment. Incremental Potential Contact (IPC) ( 22 ) is attractive in this setting because its barrier formulation, combined with continuous collision detection, robustly maintains intersection- and inversion-free simulation under large deformations and challenging contact. Recent work has improved IPC along complementary directions. For efficiency, IPC has been accelerated by projective dynamics ( 20 ) , preconditioned nonlinear conjugate gradients ( 33 ) , and GPU Gauss–Newton solvers ( 14 ; 15 ) . For accuracy, Convergent IPC ( 23 ) and GCP ( 16 ) study continuous contact potentials and their discretization. These developments have made IPC-based contact simulation increasingly useful in robotics: IPC-GraspSim ( 18 ) applies IPC to grasp simulation, while TacIPC ( 5 ) , Taccel ( 24 ) , TacEx ( 28 ) , UniVTAC ( 2 ) , and Tac2Real ( 41 ) use IPC or its variants for tactile simulation, data generation, and policy learning. Tactile sensing places a strong requirement on contact pressure accuracy, as many tactile sensors rely on contact-induced deformation to generate signals. TacEx ( 28 ) integrates GIPC ( 14 ) with Isaac Sim / Isaac Lab ( 27 ) for GelSight visuotactile sensor ( 42 ) , but GIPC-modeled objects are not mapped to high-fidelity visual assets with textures and material appearance for NVIDIA Omniverse ( 30 ) . UniVTAC ( 2 ) builds on TacEx for visuo-tactile data generation and benchmarking, and inherits this limitation. This leaves a practical gap for real-to-sim workflows that require robust deformable contact, scalable rollout, and high-fidelity rendering from detailed visual assets. In this work, we make two key contributions. • First, we introduce IsaacIPC, a simulation framework that bridges IsaacSim/Lab ( 27 ) with libuipc ( 35 ) through a dual-mesh mapper, combining high-fidelity contact simulation with realistic rendering for data collection and robot-learning policy evaluation. • Second, we propose the Geometric Mortar Contact Potential (GMCP), a mortar-based contact formulation that samples contact on tactile surfaces and incorporates the barrier potential to improve contact-pressure simulation for tactile sensing. Together, these contributions offer a practical foundation for improving contact-rich simulation in robotic interactions. 2. Related Work 2.1. Barrier-Based Contact Simulation Incremental Potential Contact (IPC) ( 22 ) formulates contact as a smoothed barrier potential with continuous collision detection (CCD), guaranteeing intersection- and inversion-free trajectories under implicit time integration. This framework has been extended along multiple axes: to rigid bodies with curved-trajectory CCD ( 10 ) , to stiff affine bodies with reduced degrees of freedom ( 19 ) , to unified rigid-deformable multibody systems with articulation constraints ( 3 ) , to high-order finite element meshes ( 9 ) , and to adaptive remeshing within the timestep ( 11 ) . A major thrust of recent work accelerates IPC to practical speeds. Lan et al. ( 20 ) integrate the barrier into projective dynamics on the GPU. Shen et al. ( 33 ) eliminate Hessian assembly via a preconditioned nonlinear conjugate gradient (PNCG) method, and Zhang et al. ( 43 ) further improve this with multilevel preconditioning. GIPC ( 14 ) derives analytic eigensystems for GPU-friendly Gauss–Newton solves, and StiffGIPC ( 15 ) adds multilevel preconditioning and affine-deformable coupling for stiff materials. Barrier-Augmented Lagrangian ( 13 ) combines barriers with augmented Lagrangian for improved conditioning, while AL-IPC ( 44 ) replaces barriers entirely with a second-order augmented Lagrangian, avoiding TOI locking. On the accuracy side, Convergent IPC ( 23 ) reformulates IPC in the continuous setting and gives a convergent discretization, with convergence demonstrated under joint refinement of the mesh, time step, and contact activation distance. The Geometric Contact Potential (GCP) ( 16 ) derives a continuum barrier from geometric interaction sets and uses an adaptive locality parameter to avoid rest-state spurious forces while relaxing the coupling between barrier extent and mesh resolution. Classical mortar contact methods improve interface consistency by enforcing contact constraints in a weak, integrated form over overlapping slave-master regions, and are widely used for accurate contact pressure transfer on nonmatching meshes ( 39 ; 31 ; 8 ) . They are typically formulated with Lagrange multiplier or augmented Lagrangian enforcement rather than feasible-iterate barrier formulations ( 16 ) ; our work instead incorporates mortar integration into a barrier potential to improve contact pressure accuracy while retaining non-penetration. 2.2. IPC in Robotic Simulation IPC has recently been adopted in robotics applications involving robust contact handling and deformable objects. IPC-GraspSim ( 18 ) introduced IPC to parallel-jaw grasp simulation with compliant fingertips, showing that explicitly modeling soft jaw deformation and frictional contact can improve grasp outcome prediction. ZeMa ( 6 ) further extended IPC-based simulation to unified rigid-deformable robotic manipulation, combining (finite element method) FEM and affine rigid bodies for applications including grasp generation, grasp repair, and reinforcement learning. Embedded IPC ( 7 ) improves the runtime of this line of work by using a reduced subspace for elasticity while retaining contact constraints on an embedded high-resolution surface, targeting interactive robot manipulation scenarios. IPC-based methods have also been explored for visuo-tactile sensing, where elastomer deformation and frictional contact directly determine the tactile signal. TacIPC ( 5 ) uses FEM with IPC to simulate optical tactile sensor gels with intersection- and inversion-free deformation, enabling tactile image and marker displacement prediction. TacEx ( 28 ) integrates GIPC ( 14 ) with Isaac Sim / Isaac Lab ( 27 ) , combining FEM with visuotactile rendering modules for GelSight sensors ( 42 ) . Taccel ( 24 ) scales this direction further by combining IPC and affine body dynamics with NVIDIA Warp ( 26 ) to support many parallel tactile-robotics environments. UniVTAC ( 2 ) builds on TacEx ( 28 ) from the data and evaluation side, providing a unified visuo-tactile simulation platform, representation-learning pipeline, and benchmark for tactile-driven manipulation. Tac2Real ( 41 ) targets online reinforcement learning and zero-shot sim-to-real transfer, using PNCG-IPC ( 33 ) as the contact solver in a