DISF: Disentangled Iterative Surface Fitting for Contact-stable Grasp Planning with Grasp Pose Alignment to the Object Center of Mass

TL;DR

This work tackles instability in surface-fitting grasp planning by adding contact-stability constraints through explicit CoM alignment. It introduces Disentangled Iterative Surface Fitting (DISF), which sequentially optimizes rotation, translation, and fingertip aperture to balance geometric compatibility with stable contact. Across 15 objects and three robot–gripper platforms in simulation, plus real-world UR3e experiments, DISF reduces CoM misalignment while maintaining surface-fit quality, yielding higher grasp success rates, especially under observed geometry. The results demonstrate improved robustness to sensor-noise and partial geometry, with clear cross-platform generalization and practical implications for reliable grasp planning from point-cloud data.

Abstract

In this work, we address the limitation of surface fitting-based grasp planning algorithm, which primarily focuses on geometric alignment between the gripper and object surface while overlooking the stability of contact point distribution, often resulting in unstable grasps due to inadequate contact configurations. To overcome this limitation, we propose a novel surface fitting algorithm that integrates contact stability while preserving geometric compatibility. Inspired by human grasping behavior, our method disentangles the grasp pose optimization into three sequential steps: (1) rotation optimization to align contact normals, (2) translation refinement to improve the alignment between the gripper frame origin and the object Center of Mass (CoM), and (3) gripper aperture adjustment to optimize contact point distribution. We validate our approach in simulation across 15 objects under both Known-shape (with clean CAD-derived dataset) and Observed-shape (with YCB object dataset) settings, including cross-platform grasp execution on three robot--gripper platforms. We further validate the method in real-world grasp experiments on a UR3e robot. Overall, DISF reduces CoM misalignment while maintaining geometric compatibility, translating into higher grasp success in both simulation and real-world execution compared to baselines. Additional videos and supplementary results are available on our project page: https://tomoya-yamanokuchi.github.io/disf-ras-project-page/

Figures (16)

• Figure 1: The relationship between the grasp planning space, geometrically aligned space, and spatially aligned space.
• Figure 2: Overview of the proposed DISF optimization process. The grasp pose optimization is disentangled into three sequential steps: (1) rotation optimization to align contact normals, (2) translation refinement for Center of Mass (CoM) alignment, and (3) gripper aperture adjustment to optimize contact point distribution. Each step iteratively updates the gripper transformation parameters to ensure both geometric compatibility and contact stability. The arrows indicate the optimization flow, illustrating how the gripper adapts to the object surface through iterative surface fitting.
• Figure 3: Contact surface optimization formulation for antipodal grasp planning. Given the canonical fingertip surfaces, we define finger contact surface patches $\mathcal{S}^f_1$ and $\mathcal{S}^f_2$ and transform them by $\mathcal{T}(\cdot;\mathbf{R},\mathbf{t},\delta d)$, where $(\mathbf{R},\mathbf{t})$ and the fingertip displacement $\delta d$ are the optimization variables. For each transformed patch, the corresponding object contact surface $\mathcal{S}^o_j$ is obtained by the correspondence function $\mathcal{H}_{\partial \mathcal{O}}(\mathcal{S}^f_j)$ on the object surface. Surface samples and normals $\{\mathbf{p}_{i_j}, \mathbf{n}^p_{i_j}\}$ and $\{\mathbf{q}_{i_j}, \mathbf{n}^q_{i_j}\}$ are used to evaluate the grasp quality $Q(\mathcal{S}^f,\mathcal{S}^o)$, which is maximized under the gripper workspace and aperture constraints.
• Figure 4: Robot--gripper platforms used in simulation for cross-platform evaluation. We evaluate grasp execution across three parallel-jaw grippers with different fingertip geometries and aperture ranges: (left) Franka Emika Panda with the Franka Hand, (middle) Universal Robots UR5e with the Robotiq HAND-E gripper, and (right) KUKA iiwa with the UMI gripper umi_gripper. The Gripper Zoom row highlights the end-effector designs to emphasize that our evaluation focuses on gripper variability rather than the robot model. The Point Cloud Representation row visualizes the corresponding canonical fingertip surfaces used in our surface-fitting optimization. In the simulator, the left fingertip is rendered as a green geom and the right fingertip as a red geom; the canonical fingertip point clouds are shown with the same color coding for consistent interpretation.
• Figure 5: Known-shape objects used in simulation, visualized in MeshLab: (a) T-shape_Block, (b) Rubber_Duck, (c) Hammer, (d) Wine_Glass, and (e) Old_Camera.