GPD: a ROS Package for Grasping Novel Objects

Dear colleagues,

We have just released our latest grasp detection software as a ROS package:

http://www.ccs.neu.edu/research/helpinghands/code.html

git page: https://github.com/atenpas/gpd

video:

The software is an easy-to-use package for grasping novel objects. Given a point cloud, the package detects 6-DOF grasp poses for a 2-finger grasp (e.g., a parallel jaw gripper). It works both on isolated objects and in densely cluttered scenes, and it has already been deployed on various platforms, including a UR5, Baxter, HSR, …

Features:

  • No object models required
  • No object segmentation required
  • No training required
  • Grasp poses are 6-DOF (allows a variety of grasp approach directions)
  • Can use any point cloud (better point clouds results in better performance)
  • Visualization in rviz or pcl

Requirements:

  • RGBD camera, Nvidia graphics card with CUDA
  • Caffe, PCL 1.7 or higher, Eigen
  • Ubuntu 14.04 with ROS Indigo or Ubuntu 16.04 with ROS Kinetic

Check it out!

Andreas ten Pas and Rob Platt
Helping Hands Lab
College of Computer and Information Science
Northeastern University

3 Likes