GraspQP: Differentiable Optimization of Force Closure for Diverse and Robust Dexterous Grasping

2025-08-20

AI Summary

Existing grasp dataset generation methods for dexterous hands predominantly rely on sampling or simplified force-closure analysis, resulting in limited grasp diversity, poor physical feasibility, and inadequate coverage of complex manipulations such as pinch grasps and three-finger precision grasps. To address these limitations, we propose a differentiable force-closure energy modeling framework, enabling end-to-end optimization via implicit quadratic programming. We further introduce a dynamic gradient rejection mechanism to enhance the MALA* sampler, significantly improving optimization efficiency and grasp diversity. Leveraging large-scale parallel computation, we generate a high-quality, large-scale dataset across 5,700 objects from DexGraspNet, encompassing five gripper types and three grasp modalities. Experiments demonstrate that our method outperforms existing benchmarks in both grasp stability and diversity, providing robust, physically plausible data to support training of dexterous manipulation models and task-level planning.

Technology Category

• Intelligent Robots: Manipulation
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Learning with Manifolds

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Web Mining and Content Analysis: Web data generation and simulation

Abstract

Dexterous robotic hands enable versatile interactions due to the flexibility and adaptability of multi-fingered designs, allowing for a wide range of task-specific grasp configurations in diverse environments. However, to fully exploit the capabilities of dexterous hands, access to diverse and high-quality grasp data is essential -- whether for developing grasp prediction models from point clouds, training manipulation policies, or supporting high-level task planning with broader action options. Existing approaches for dataset generation typically rely on sampling-based algorithms or simplified force-closure analysis, which tend to converge to power grasps and often exhibit limited diversity. In this work, we propose a method to synthesize large-scale, diverse, and physically feasible grasps that extend beyond simple power grasps to include refined manipulations, such as pinches and tri-finger precision grasps. We introduce a rigorous, differentiable energy formulation of force closure, implicitly defined through a Quadratic Program (QP). Additionally, we present an adjusted optimization method (MALA*) that improves performance by dynamically rejecting gradient steps based on the distribution of energy values across all samples. We extensively evaluate our approach and demonstrate significant improvements in both grasp diversity and the stability of final grasp predictions. Finally, we provide a new, large-scale grasp dataset for 5,700 objects from DexGraspNet, comprising five different grippers and three distinct grasp types. Dataset and Code:https://graspqp.github.io/

Problem

• Generating diverse and physically feasible dexterous grasps beyond power grasps
• Overcoming limitations of sampling-based algorithms and simplified force-closure analysis
• Providing large-scale grasp data for manipulation policies and task planning

Innovation

• Differentiable energy formulation via Quadratic Program
• Adjusted optimization method MALA* for performance
• Large-scale diverse grasp synthesis beyond power grasps