GraspSampler

GraspSampler is a library that performs efficient sampling of millions of parallel-jaw grasps around a target object object. One can quickly generate a dataset that contains numerous grasps and their quality scores. The project GraspFlow used GraspSampler to sample numerous grasps and labelled them in the simulation framework IsaacGymGrasp.

This library supports visualization of grasps, grippers and objects, rendering of 3D point clouds, and efficient sampling of parallel-jaw grasps. The tutorials show examples using a Franka-Emika Gripper and a target object.

<!-- Grasp Definition -->

Installation

GraspSampler was created using Ubuntu 18.04 and python 3.8. To install GraspSampler on your local machine follow these intstructions.

Once you have a working environment for GraspSampler, you can try the tutorials in the next section.

Tutorials

The following are tutorials on how to use the different features of GraspSampler.

Tutorial 1: Gripper Definitiion

This tutorial shows how to load a gripper into the scene and manipulate its position. Run the gripper tutorial:

python -m tutorials.gripper

Tutorial 2: Adding Target Object

This tutorial shows how to add a target object into the scene. Run the target object tutorial:

python -m tutorials.object

Tutorial 3: Visualizing Object Point Clouds

This tutorial shows how to generate a point cloud from an object in the scene. Run the point cloud tutorial:

python -m tutorials.pc_manager

Tutorial 4: Sampling Grasps

This tutorial loads an object and samples multiple grasps about that object. Run the grasp sampler tutorial:

python -m tutorials.graspsampler

<!-- ## XXX -->