Abstract
Accurately simulating whether an object will be lifted securely or dropped during grasping is a longstanding Sim2Real challenge. Soft compliant jaw tips are almost universally used with parallel-jaw robot grippers due to their ability to increase contact area and friction between the jaws and the object to be manipulated. However, interactions between the compliant surfaces and rigid objects are notoriously difficult to model. We introduce IPC-GraspSim, a novel grasp simulator that extends Incremental Potential Contact (IPC) - a highly accurate collision + deformation model developed in 2020 for computer graphics. IPC-GraspSim models both the dynamics and the deformation of compliant jaw tips to reduce the Sim2Real gap for robot grasping. We evaluate IPC-GraspSim using a set of 2,000 physical grasps across 16 adversarial objects where analytic models perform poorly. In comparison to both analytic quasistatic contact models (soft point contact, REACH, 6DFC) and dynamic grasp simulators (Isaac Gym with FleX), results suggest IPC-GraspSim can predict robustness with higher precision and recall (F1 = 0.85). IPC-GraspSim increases F1 score by 0.03 to 0.20 over analytic baselines and 0.09 over Isaac Gym, at a cost of 8000x and 1.5x more compute time, respectively.
Introduction
Bridging the "Sim2Real gap"—the performance difference between simulated environments and the physical world—remains a critical hurdle in robotics. In the context of parallel-jaw grasping, this gap is largely attributed to the behavior of soft gripper pads. While these compliant tips are essential for increasing friction and contact surface area, their deformation is difficult to simulate accurately. Traditional simulators often rely on penalty-based contact methods that allow for unrealistic interpenetration or "force-field" effects, leading to false positives where a grasp appears stable in simulation but fails in reality.
Methodology
To address these limitations, we present IPC-GraspSim, which replaces standard collision engines with the Incremental Potential Contact (IPC) model. IPC provides an intersection-free guarantee through a smoothed barrier method and continuous collision detection, ensuring that surfaces never interpenetrate regardless of the applied forces.
The simulator employs a Neo-Hookean material model to capture the non-linear elastodynamics of the gripper pads. The grasping logic is implemented as a state machine: the jaws close until a specific deformation energy threshold is reached, the object is lifted, and success is determined by the final relative pose of the object.
Experimental Setup
We evaluated IPC-GraspSim using a physical dataset of 2,000 grasp attempts performed by an ABB YuMi robot. The dataset features 16 "adversarial" 3D-printed objects with complex geometries that typically cause analytic models to fail. Three objects were utilized to calibrate the simulator's material parameters—specifically Young’s modulus ($E$) and the friction coefficient ($\mu$)—while the remaining 13 objects were used for the final performance evaluation.
Results and Discussion
IPC-GraspSim achieved a superior F1 score of 0.85, outperforming analytic baselines such as REACH and 6DFC (0.65–0.82) and dynamic simulators like NVIDIA’s Isaac Gym (0.76). A significant factor in this success is the simulator's ability to model the simultaneous dynamics and compliance of the grasp; it accurately captures how an object rotates and slides as the gripper pad conforms to its geometry.
Furthermore, IPC-GraspSim eliminates "false sticking," a common numerical artifact in other simulators where objects remain unnaturally attached to the gripper. Our findings also indicate that high-fidelity modeling of gripper geometry (e.g., tip curvature) is vital for accurate Sim2Real transfer.
The primary trade-off for this accuracy is computational cost. IPC-GraspSim requires approximately 605 seconds per grasp, which is 1.5x slower than Isaac Gym and significantly slower than analytic models. However, this level of fidelity is highly valuable for generating high-quality offline datasets for training.
Conclusion and Future Work
IPC-GraspSim demonstrates that integrating advanced contact models like IPC can significantly reduce the Sim2Real gap for robotic manipulation. Future research will explore speed optimizations, such as the "Medial-IPC" method, and incorporate structural bending effects (e.g., finger cantilever) to further refine prediction accuracy.
Summary updated