Force-Aware 3D Contact Modeling for Stable Grasp Generation

TL;DR

This paper defines a force-aware contact representation and introduces force-aware stability constraints that can help identify key contact points for stability which provide effective initialization and guidance for optimization towards a stable grasp.

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.

Figures (6)

• Figure 1: The difference between previous methods (without force predictions) and ours (with force predictions). Force information guides the optimization by identifying a few keypoints to ensure stability.
• Figure 2: Our framework consists of a generator and an optimizer. The object features (OF) are obtained by processing sampled object points using PointNet++. Then the generator consisting of 3 successive cVAEs takes OF as an input condition. It generates force-aware contacts including the contact normal force value, which are then used to recognize important keypoints for equilibrium. In optimization stage, I. and II. are two-stage initialization based only on keypoints, and III. optimizes MANO parameters $\boldsymbol{\beta}, \boldsymbol{\theta}$ while also being guided by the keypoints to get the stable grasp pose.
• Figure 3: The friction cone (red) and its approximation (blue) that is used in formulating the stability energy function
• Figure 4: Grasp Samples from both datasets. Each sample is shown in two views with highlights of penetrations, possible supportive forces at the contact points, and gravity labels.
• Figure 5: Normal contact forces by our method, avg. predictor, and GT