PointClud vision pipeline detects robot body as a part, instead of the actual box

asked 2019-03-11 02:27:12 -0600

artemiialessandrini gravatar image


Please, refer to the video's right side: pcl methods detect the box and I set some raw coordinate frame on it, so arm could move towards.

(There are some UR5 PID controllers issues in the left side, I dunno a solution, but it's not a question for now..)

By current code, arm tries to move to the detected box coordinate frame, but at some moment it detects the robot body itself to be a part and application fails. I'm getting the biggest cluster on the table.

I need to kinect only detect an actual box.

I believe, there's a way to stop detection node, while move_arm_one starts, but cannot get how to implement it.

Feel free to ask any questions! Thanks for your help in advance!

The code:

def move_arm_one():

pose_goal = geometry_msgs.msg.Pose()
pose_goal.orientation = orient
pose_goal.position.x = x
pose_goal.position.y = y
pose_goal.position.z = z + 0.2

roll, pitch, yaw = t.euler_from_quaternion(orient)
newroll = roll + pi # putting the hand in the inverse position
newquat = t.quaternion_from_euler(newroll, pitch, yaw)
pose_goal.orientation = Quaternion(newquat[0], newquat[1], newquat[2], newquat[3])

plan1 = arm_group.plan()

# rospy.sleep(5)


# rospy.sleep(5)
# arm_group.set_start_state_to_current_state()
rospy.loginfo( " Planning arm init: Succeed")
print pose_goal

arm_group.go(pose_goal, wait=True)
# rospy.sleep(1) 

if __name__ == '__main__':

display_trajectory_publisher = rospy.Publisher('/move_group/display_planned_path',
rospy.init_node('moveit_grasp', anonymous=True)
rospy.loginfo("Starting grasp app")

robot = moveit_commander.RobotCommander()
scene = moveit_commander.PlanningSceneInterface()
arm_group = moveit_commander.MoveGroupCommander("manipulator")
gripper_group = moveit_commander.MoveGroupCommander("gripper")


arm_group.set_planer_id = "RRTkConfigDefault"

planning_frame = arm_group.get_planning_frame() # reference frame for robot
print " Reference frame: %s" % planning_frame
eef_link = arm_group.get_end_effector_link() # end-effector link
print " End effector link: %s" % eef_link
group_names = robot.get_group_names() # all the groups in the robot
print " Available Robot Groups:", robot.get_group_names()

# creating a tf listener
listener = tf.TransformListener()
rate = rospy.sleep(5.0) # filling up a tf buffer to localize a tfansform
rospy.loginfo("Filling up a buffer...")
    ((x, y, z), orient) = listener.lookupTransform('/part', '/camera_rgb_optical_frame', rospy.Time(0))
    rospy.loginfo("Position %s; Orientation %s", (x, y, z), orient)
except (tf.LookupException, tf.ConnectivityException, tf.ExtrapolationException):
    rospy.logerr("Transform failed.")

# add movements

moveit_commander.roscpp_shutdown() # roscpp_shutdown()
edit retag flag offensive close merge delete


That's not all the code. The code you posted doesn't contain your (custom?) detection node, which you say you can't figure out how to turn on and off. Can you try explaining again with more precision and detail? You are "getting the biggest cluster on the table"? Do you mean your perception pipeline assumes that a box of certain dimensions is where it sees the most points? What do you mean with "I need to kinect only detect an actual box"?

fvd gravatar image fvd  ( 2020-03-04 07:10:47 -0600 )edit