Force-Aware 3D Contact Modeling
for Stable Grasp Generation
Abstract
Contact-based grasp generation plays a crucial role in various applications. Recent methods typically focus on the geometric structure of objects, producing grasps with diverse hand poses and plausible contact points. However, these approaches often overlook the physical attributes of the grasp, specifically the contact force, leading to reduced stability of the grasp. In this paper, we focus on stable grasp generation using explicit contact force predictions. First, we define a force-aware contact representation by transforming the normal force value into discrete levels and encoding it using a one-hot vector. Next, we introduce force-aware stability constraints. We define the stability problem as an acceleration minimization task and explicitly relate stability with contact geometry by formulating the underlying physical constraints. Finally, we present a pose optimizer that systematically integrates our contact representation and stability constraints to enable stable grasp generation. We show that these constraints can help identify key contact points for stability which provide effective initialization and guidance for optimization towards a stable grasp. Experiments are carried out on two public benchmarks, showing that our method brings about 20% improvement in stability metrics and adapts well to novel objects.
Our method jointly capture the distribution of contact likelihood and force values, which are then used to extract key contact points using a simple yet effective physics model. The keypoints provide strong initialization and guidance for grasp optimization.
The labelled contact forces using our automatic labelling pipeline.
The generated human grasps, compared with previous state-of-the-art methods.
Quantitative comparison on two public benchmarks. Our method outperforms previous methods in terms of stability metrics by a large margin.
Incorporating contact forces in human grasp generation
The relationship between grasping geometry and contact forces on general objects is currently under-explored. Meanwhile, we find that if the contact forces can be predicted, then modeling stability in contact-based grasp synthesis will be much easier. With explicit contact forces predicted, we can 1) explicitly formulate stability; 2) identify key contact points that are most critical for stability. Thus, we focus on modeling the contact forces to facilitate stable grasp generation.
Automatic Contact Force Labelling
We propose an automatic contact force labelling pipeline to label the contact forces on hand-object interaction datasets. Given a human-object interaction sample, we first do convex decomposition separately. Then, we set up a physics simulation environment to simulate the grasping process and record the contact forces on each contact point. Finally, we transform the continuous normal force values into discrete levels and encode them using one-hot vectors to form our force-aware contact representation.
Stability-Guided Keypoint Selection and Keypoint-Guided Grasp Optimization
We relate physical stability with geometric properties by identifying keypoints. As few as 3 points can form a stable grasp according to force closure theory. Thus, we intuitively formulate the stability measure as the magnitude of accelerations, based on which the keypoint set with the least acceleration is seleted. These keypoints provide effective initialization and guidance for optimization towards a stable grasp. We formulate a keypoint-guided optimization framework that systematically integrates our force-aware contact representation and stability constraints to generate stable grasps.
State-of-the-art Stability Performance
We conduct experiments on two public benchmarks, GRAB for in-domain test and HO3D for out-of-domain test, showing that our method outperforms previous methods by a large margin (around 20%) in terms of stability metrics. We also measure the success rate which considers both stability and penetration, and our method achieves a significant improvement over previous methods. Extensive ablation studies are available in the paper.
BibTeX
@inproceedings{chen2026forceaware3d,
title={Force-Aware 3D Contact Modeling for Stable Grasp Generation},
author={Chen, Zhuo and Zhang, Zhongqun and Cheng, Yihua and Leonardis, Ales and Chang, Hyung Jin},
booktitle={Annual AAAI Conference on Artificial Intelligence (AAAI)},
year={2026},
}