[Paper Review] Grasp Planning for Customized Grippers by Iterative Surface Fitting

Yongxiang Fan, Hsien-Chung Lin+2 more|arXiv (Cornell University)|Mar 30, 2018

Robot Manipulation and Learning3 references3 citations

TL;DR

This paper proposes an Iterative Surface Fitting (ISF) algorithm for real-time grasp planning of customized grippers by simultaneously optimizing palm pose and finger displacement to minimize surface fitting error. Guided sampling avoids local optima, enabling robust grasps on complex, cluttered objects using unsegmented point clouds with an average planning time under 0.1 seconds.

ABSTRACT

Customized grippers have broad applications in industrial assembly lines. Compared with general parallel grippers, the customized grippers have specifically designed fingers to increase the contact area with the workpieces and improve the grasp robustness. However, grasp planning for customized grippers is challenging due to the object variations, surface contacts and structural constraints of the grippers. In this paper, an iterative surface fitting (ISF) algorithm is proposed to plan grasps for customized grippers. ISF simultaneously searches for optimal gripper transformation and finger displacement by minimizing the surface fitting error. A guided sampling is introduced to avoid ISF getting stuck in local optima and improve the collision avoidance performance. The proposed algorithm is able to consider the structural constraints of the gripper and plan optimal grasps in real-time. The effectiveness of the algorithm is verified by both simulations and experiments. The experimental videos are available at: http://me.berkeley.edu/\%7Eyongxiangfan/CASE2018/caseisf.html

Motivation & Objective

• Address the challenge of grasp planning for customized grippers with large surface contact and structural constraints.
• Improve grasp robustness and stability by leveraging surface matching between gripper and object geometry.
• Enable real-time grasp planning despite complex object shapes and cluttered environments.
• Overcome limitations of traditional point-contact models and computationally heavy optimization in surface-based grasp planning.
• Integrate structural constraints (e.g., jaw width, DOFs) and contact normal alignment into the grasp planning process.

Proposed method

• Propose an Iterative Palm-Finger Optimization (IPFO) framework that jointly optimizes palm pose and finger displacement using closed-form solutions.
• Define a surface fitting error metric based on distance between gripper and object surfaces and misalignment of contact normals.
• Implement guided sampling to initialize ISF with high-quality grasp candidates, reducing risk of local optima.
• Use unsegmented 3D point clouds from stereo cameras as input, with surface smoothing and normal estimation for real-world robustness.
• Apply iterative optimization to minimize surface fitting error across multiple iterations until convergence.
• Incorporate structural constraints (e.g., jaw width, DOFs) directly into the optimization to ensure feasible gripper configurations.

Experimental results

Research questions

• RQ1 How can grasp planning for customized grippers be made efficient and robust when considering large surface contacts and structural constraints?
• RQ2 Can iterative surface fitting with joint palm-finger optimization achieve real-time performance in complex environments?
• RQ3 How does guided sampling improve convergence and collision avoidance in surface-based grasp planning?
• RQ4 To what extent can the algorithm handle unsegmented, noisy point clouds in cluttered industrial settings?
• RQ5 Can the method generalize to objects with complex geometries and varying sizes without retraining?

Key findings

• The ISF algorithm achieved an average grasp planning time of 64.4 ms per collision-free grasp in simulations, with real-time performance confirmed by experiments.
• In simulations, the algorithm successfully found 36.2 collision-free grasps per 60 samples on average across nine test objects.
• The guided sampling strategy significantly improved convergence and reduced the risk of local optima, as evidenced by stable fitting error reduction from 10 mm to 1 mm in 7 iterations.
• Experiments on a FANUC industrial robot demonstrated successful grasp execution on six diverse objects, including complex shapes like Doraemon and screwdrivers.
• In a heavy clutter environment with unsegmented point clouds, the algorithm successfully identified and executed sequential grasps despite occlusions and complex surface compositions.
• The surface fitting error was reduced from 10 mm to 1 mm within 7 iterations of IPFO, confirming effective convergence and surface matching.