Abstract

Synthesizing 3D whole bodies that realistically grasp objects is crucial for animation, mixed reality, and robotics.
Key challenges include natural coordination between hand, body, and environment, and the scarcity of training data.
CWGrasp overcomes these by performing geometry-based reasoning early: we sample a ReachingField (a probabilistic direction field around the object), then condition both the grasping hand (CGrasp) and the reaching body (CReach) on this direction, and finally run a lightweight optimization to resolve penetrations.
CWGrasp handles both left- and right-hand grasps, runs ~16× faster than exhaustive baselines (e.g., FLEX), and outperforms them on GRAB and ReplicaGrasp datasets.
Code and models: https://gpaschalidis.github.io/cwgrasp.

Teaser

3DV 2025

We develop **CWGrasp**, a framework for synthesizing 3D whole-body grasps on objects placed on receptacles. By integrating early geometry-based reasoning (ReachingField) with controllable synthesis, we achieve realistic grasps at a fraction of the cost compared to prior art.

Method Overview

1. ReachingField: Ray-cast from the object to build a probabilistic direction field.
2. CGrasp & CReach: Generate a hand and a body guided by the sampled direction.
3. Optimization: Resolve penetrations with a fast joint refinement.

ReachingField

CGrasp & CReach

CGrasp: Controllable hand synthesis following the direction.

CReach: Directional body reaching synthesis.

Optimization Comparison

Qualitative Results: Right-Hand

Low-height Object

Medium-height Object

High-height Object

Qualitative Results: Left-Hand

BibTeX

@inproceedings{paschalidis2025cwgrasp,
  title     = {{3D} Whole-Body Grasp Synthesis with Directional Controllability},
  author    = {Paschalidis, Georgios and Wilschut, Romana and Antić, Dimitrije and Taheri, Omid and Tzionas, Dimitrios},
  booktitle = {International Conference on 3D Vision (3DV)},
  year      = {2025},
}