HUG

Human grasp inference for manipulation.

Access Required

This package requires private repository access.

Repository: HUG (private)

Purpose

Grasp inference from human demonstrations:

• Grasp pose prediction

• Object manipulation

• Learning from observation

Installation

HUG is a submodule of lerobot_ros2:

cd lerobot_ros2
git submodule update --init HUG
cd HUG
uv sync  # or pip install -e .

Usage

from hug import GraspInference

model = GraspInference.load("path/to/model")
grasp_pose = model.predict(rgb_image, depth_image)

Integration

Used with lerobot_ros2 for:

• Data collection with grasp labels

• Policy training

• Deployment

Notes

• No training secrets in public docs

• Model weights separate from code

• See internal documentation for training

Related

• lerobot_ros2