6 DoF grasp synthesis project

Three ROS packages have been developed to implement two different solutions to grasping:

1. contact_graspnet - ROS Wrapper of Contact-Graspnet with modifications made to switch from a Panda gripper to the TIAGo gripper. This method uses a learning based approach.

2. geometric_grasp - A ROS package with an analytical approach to grasping.

3. pick_up_object - State machine that allows for the execution of both grasping methods.

Prerequisites

Kinetic docker container.

Melodic docker container with TensorFlow and Nvidia support. Refer to tiago_docker_tensorflow.

How to run it

1. Choose what launch file to launch depending on what method you would like to test:

  `roslaunch contact_graspnet contact_graspnet.launch`
  
  `roslaunch geometric_grasp geometric_grasp.launch`

2. roslaunch pick_up_object pick_up_object.launch

3. In order to switch the different methods of grasping, it must be addressed in the generate_grasps state.

Pick up pipeline

Overview of the full pipeline to grasp an object.

Image demonstration of Contact-GraspNet inference

Image demonstration of the Analytical approach grasps prediction