Pickup experiment with GPD

Description

ROS package for implementing the grasping algorithms GPD (Grasp Pose Detection) on Kinova Gen3 7DOF.

All works were done on Ubuntu 20.04 with ROS Noetic.

Required ROS packages for the Gen3:

Software and version

• ros_kortex: any
• ros_kortex_vision: any
• gpd_ros: 2.0.0
• opencv: 4.2.0
• PCL: 1.10

Actions for grasping task

• motionPlanning
  • getCurrentState
  • moveToJointConfig
    • rotateSingleJoint
    • moveToInitial
  • moveToCartesianPose
    • moveToPreGraspPose
    • moveToGraspPose
    • moveToPostGraspPose
  • gripperCommand
    • openGripper
    • closeGripper
    • openToGripperPosition
• perception
  • camera
    • takeSnapshot
  • pcl
    • passThroughFilter
    • planarSegmentation
    • voxelGrid
    • concatenatePointCloud
• GraspPose
  • generateGraspPose

Note:

• When ticking moveToJointConfig and moveToCartesianPose, both actions callback getCurrentState to obtain current joint configuration (joint angles for each joint, i.e., $q_1$,...,$q_7$) and current Cartesian pose (position and orientation, i.e., x, y, z, $\theta_x$, $\theta_y$, $\theta_z$, of the end-effector).

TODO

• determine parameters ports, both in and out, for each action.

Task sequence (workflow) for a single grasping task

• openGripper
• moveToInitial
• takeSnapShot
• passThroughFilter
• planarSegmentation
• voxelGrid
• generateGraspPose
• moveToPreGraspPose
• moveToGraspPose
• closeGripper
• moveToPreGraspPose
• moveToPostGraspPose