🕷️Human Universal Grasping🕷️ 👉HUG is a flow-matching model that generates diverse...

2026-06-19 · Source: AI with Papers - Artificial Intelligence & Deep Learning (@AI_DeepLearning) - Telegram · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Computer Vision · Depth: Expert, quick

Summary

Human Universal Grasping (HUG) is introduced as a novel flow-matching model designed to generate diverse human grasps. This model can produce a variety of natural-looking hand configurations for any object specified by a user. Its key capability lies in processing a single RGB-D image, captured from a stereo camera, to infer and generate these complex grasping poses. This approach simplifies the input requirements for robotic grasping systems, moving beyond traditional methods that often require detailed 3D models or multiple views. HUG's ability to generate diverse grasps suggests a robust understanding of human-like interaction with objects, making it suitable for applications requiring adaptable and natural manipulation. The project includes a paper, repository, and project page for further details.

Key takeaway

For Robotics Engineers developing manipulation systems, HUG offers a significant advancement by enabling diverse, human-like grasping from minimal input. You should consider integrating flow-matching models like HUG to enhance the adaptability and naturalness of your robotic grippers, especially in environments where detailed 3D object models are unavailable. This could streamline deployment and improve interaction with novel objects.

Key insights

HUG is a flow-matching model generating diverse human grasps from a single RGB-D image.

Method

HUG employs a flow-matching model to synthesize diverse human grasps from a single RGB-D image input, enabling versatile object interaction.

In practice

• Generate varied human-like grasps for robotics.
• Utilize single RGB-D camera input for grasping.

Topics

• Human Universal Grasping
• Flow-matching Models
• Robotic Grasping
• RGB-D Imaging
• Computer Vision
• Human-Robot Interaction

Code references

• KevinyWu/hug

Best for: Research Scientist, AI Scientist, Robotics Engineer, Computer Vision Engineer

Related on AIssential

• EAGG: Embodiment-Aligned Grasp Generation via Geometry-Aware Graph Conditioning — Aligning embodiment structure within a shared generator strengthens cross-end-effector grasp generation.
• Generating Natural and Expressive Robot Gestures through Iterative Reinforcement Learning with Human Feedback using LLMs — Iterative RLHF with human feedback enhances LLM-generated robot gestures for natural and expressive human-robot interaction.
• UniDexTok: A Unified Dexterous Hand Tokenizer from Real Data — UniDexTok unifies heterogeneous dexterous hand data into a shared discrete token space, enabling accurate cross-embodiment state representation and transfer.
• What Matters When Cotraining Robot Manipulation Policies on Everyday Human Videos? — Cotraining robot policies with everyday human videos requires high-quality hand pose data and specialized networks to bridge the motion gap.
• DragMesh-2: Physically Plausible Dexterous Hand-Object Interaction with Articulated Objects — DragMesh-2 enables robust dexterous interaction with articulated objects by modeling contact dynamics and injecting physical signals into policy learning.
• Beyond Binary: Sim-to-Real Dexterous Manipulation with Physics-Grounded Contact Representation — Center-of-Pressure (CoP) enables robust sim-to-real dexterous manipulation by preserving dense, physics-grounded tactile information.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI with Papers - Artificial Intelligence & Deep Learning (@AI_DeepLearning) - Telegram.