pr2_eih_ros

register hand kinect

to run:

roslaunch kinect_transform ar_hand.launch
roslaunch kinect_transform carmine_registration.launch
rosrun kinect_transform kinect_transform

kinfu

to run:

roslaunch kinect_transform wrist_to_hand.launch
roslaunch pcl_utils kinfu.launch
roslaunch pcl_utils occlusion_parameters.launch
rosrun pcl_utils kinfu
rostopic pub /get_occlusions std_msgs/Empty -1 [this will make kinfu download its data and perform occlusion finding once]

publishes:

/kinfu/camera_points (current [hand] camera points)
/kinfu/kinfu_points (zero-crossing points downloaded from kinfu)
/kinfu/objects (visualization markers - faces and gaussians)
/kinfu/occluded_points (clusters and points behind them)
/kinfu/occluded_region_array (faces and gaussians)

to be added:

/kinfu/graspable_points [all points above table]
/kinfu/plane_bounding_box

handle detector

input:

/kinfu/graspable_points

to run:

roslaunch handle_detector kinfu_handle_detector
rosrun handle_poses_publisher

publishes:

/handle_detector/avg_handle_poses

bsp:

outline:

1. get ___ closest regions from kinfu
2. initialize BSP optimization trajectory
3. run bsp
4. execute controls
5. if no good handles to grasp: 6. go to 1
6. else: 8. trajectory <- /check_handle_grasps/trajectory [on callback] 9. execute grasp a. move in b. close gripper c. move up d. move back e. drop off

to run:

(TBD)

grasping:

input:

/handle_detector/avg_handle_poses
/kinfu/graspable_points
/kinfu/plane_bounding_box

outline:

1. convex decomposition of pointcloud - add table from table pose
2. for each handle pose:
   1. call trajopt (with collisions)
   2. if no collision && reaches handle:
   3. publish pose trajectory [/check_handle_grasps/trajectory]

publishes:

/check_handle_grasps/trajectory