Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EFF-Grasp: Energy-Field Flow Matching for Physics-Aware Dexterous Grasp Generation

Yukun Zhao, Zichen Zhong, Yongshun Gong, Yilong Yin, Haoliang Sun

Abstract

Denoising generative models have recently become the dominant paradigm for dexterous grasp generation, owing to their ability to model complex grasp distributions from large-scale data. However, existing diffusion-based methods typically formulate generation as a stochastic differential equation (SDE), which often requires many sequential denoising steps and introduces trajectory instability that can lead to physically infeasible grasps. In this paper, we propose EFF-Grasp, a novel Flow-Matching-based framework for physics-aware dexterous grasp generation. Specifically, we reformulate grasp synthesis as a deterministic ordinary differential equation (ODE) process, which enables efficient and stable generation through smooth probability flows. To further enforce physical feasibility, we introduce a training-free physics-aware energy guidance strategy. Our method defines an energy-guided target distribution using adapted explicit physical energy functions that capture key grasp constraints, and estimates the corresponding guidance term via a local Monte Carlo approximation during inference. In this way, EFF-Grasp dynamically steers the generation trajectory toward physically feasible regions without requiring additional physics-based training or simulation feedback. Extensive experiments on five benchmark datasets show that EFF-Grasp achieves superior performance in grasp quality and physical feasibility, while requiring substantially fewer sampling steps than diffusion-based baselines.

EFF-Grasp: Energy-Field Flow Matching for Physics-Aware Dexterous Grasp Generation

Abstract

Denoising generative models have recently become the dominant paradigm for dexterous grasp generation, owing to their ability to model complex grasp distributions from large-scale data. However, existing diffusion-based methods typically formulate generation as a stochastic differential equation (SDE), which often requires many sequential denoising steps and introduces trajectory instability that can lead to physically infeasible grasps. In this paper, we propose EFF-Grasp, a novel Flow-Matching-based framework for physics-aware dexterous grasp generation. Specifically, we reformulate grasp synthesis as a deterministic ordinary differential equation (ODE) process, which enables efficient and stable generation through smooth probability flows. To further enforce physical feasibility, we introduce a training-free physics-aware energy guidance strategy. Our method defines an energy-guided target distribution using adapted explicit physical energy functions that capture key grasp constraints, and estimates the corresponding guidance term via a local Monte Carlo approximation during inference. In this way, EFF-Grasp dynamically steers the generation trajectory toward physically feasible regions without requiring additional physics-based training or simulation feedback. Extensive experiments on five benchmark datasets show that EFF-Grasp achieves superior performance in grasp quality and physical feasibility, while requiring substantially fewer sampling steps than diffusion-based baselines.
Paper Structure (32 sections, 18 equations, 7 figures, 8 tables)

This paper contains 32 sections, 18 equations, 7 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Dexterous Grasp Generation via Flow Matching
  4. Physical-Aware Sampling via Energy Field Flow
  5. Explicit Physical Energy Fuctions.
  6. Local Monte Carlo Estimation for $g_t$.
  7. Experiments
  8. Experimental Setup
  9. Datasets.
  10. Evaluation Metrics.
  11. Baselines.
  12. Comparative Analysis of Grasp Generation Performance
  13. Overall Performance and Physical Realism.
  14. Stability vs. Diversity Analysis.
  15. Efficiency and Quality Analysis
  16. ...and 17 more sections

Figures (7)

  • Figure 1: Visualization of the physics-aware generation process in EFF-Grasp.
  • Figure 2: Overview of the EFF-Grasp sampling process. The method integrates Flow Matching with the training-free physics-aware guidance strategy. By evaluating candidate grasps using energy fields via local Monte Carlo sampling, we compute the physical energy field to guide the ODE trajectory towards physically feasible regions.
  • Figure 3: Visualization of generation results produced by our method across all datasets.
  • Figure 4: Compared to the SOTA method DGA, EFF-Grasp generates more stable and physically feasible grasps (see the boxed failure cases of DGA).
  • Figure 5: Comparison of Success Rate and Penetration Depth versus NFE on MultiDex and RealDex datasets. The left y-axis represents the Success Rate (%), and the right y-axis indicates the Penetration Depth (mm). Solid lines denote our method, while dashed lines represent the DGA baseline. Our method consistently outperforms DGA, achieving significantly higher success rates and lower penetration depths, particularly demonstrating faster convergence at lower NFEs.
  • ...and 2 more figures