ur10e_controller

Robot control, realtime reconstruction, and defect detection for AI4CE lab M3DP project.

Author: Yongyu Liu [email protected]

Tutorial

Hardware connection and basic setup

For more information, check official repository Universal_Robots_ROS_Driver

1. Ensure that the computer has enabled real-time capablities, run uname -a and find PREEMPT_RT. This is necessary for stable connection.

   

   

2. Wired connect computer with the robot arm, in computer network settings, find the wired connection and set it as in the picture.

   

3. On robot arm's teaching pendant, find Settings -> System -> Network and set it.

   

4. Start roscore

5. In robot's NORMAL mode, load an external control program

6. Start a ROS node that contains $(find ur_robot_driver)/launch/ur10e_bringup.launch

7. Start the loaded program on the teaching pendant.

You should see the following message in the driver node

Hand-eye calibration

For more information, check easy_handeye

1. Run roslaunch ur10e_controller handeye_calibration.launch, once 3 windows pop up, the errors can be safely ignored.

2. Run rqt. In Plugins -> Visualization -> Image view, subscribe topic /aruco_tracker/result. You should see these windows.

3. Only take samples when the result shown in rqt is correct and stable. The calibration result will be stored at $HOME/.ros/easy_handeye/ur10e_calib_eye_on_hand.yaml

Experiment

1. Run roslaunch ur10e_controller bringup_all.launch.

2. Run roslaunch ur10e_controller publish_calibration.launch, check that frame camera_color_optical_frame has appeared in RViz.

3. Run rosrun ur10e_controller realtime.py, a window displaying real-time point clouds after denoise and reconstruction will pop up.

   Whenever a message is published in the topic /height, by printing program or via terminal, e.g. rostopic pub /height std_msgs/Float64 0.04, the window will make a comparison between current scene and model cropped in z-axis by corresponding height.

4. Run rosrun ur10e_controller print_reconstruct.py, the robot will compute and store plans for printing a cubic, then waits input in terminal to execute the plans.

   Or run rosrun ur10e_controller direct_print.py, which loads pre-computed plans and execute them.

5. Run python3 scripts/printhead_control.py to manually control the printhead motor, or uncomment arduino lines in print_reconstruct.py or direct_print.py for automatic control and the motor will be stopped when the robot is rotating.

Troubleshoot

Do not mess with camera's calibration. You can try More -> Calibration Data -> Restore Factory in Intel RealSense Viewer SDK, but do not try other self-calibration methods in SDK unless you have a good reason and are familiar with depth camera calibration. This may result in wrong hand-eye calibration result and erroneous point cloud reconstruction.

For scripts that contain processing point clouds, check the variable depth_scale = profile.get_device().first_depth_sensor().get_depth_scale(). Due to unknown hardware issues, the value of depth_scale is either 9.999999747378752e-05 or 0.0010000000474974513. Although they refer to the same length unit, the point cloud will be scaled incorrectly when depth_scale < 0.001. The solution is to scale the point cloud down by 0.1 once it is created from rgbd image, e.g. in realtime.py.

If the error Cannot open /dev/ttyACM0: Permission denied occurs when connecting to the arduino board, run sudo chmod a+rw /dev/ttyACM0 and retry.

There are parameters to be tuned:

• Print: length, height, fluid_width, layers
• Capture frame rate: dt
• Denoise: bbox = o3d.geometry.AxisAlignedBoundingBox(min_bound=(-fluid_width, -fluid_width, -fluid_width), max_bound=(length+fluid_width, length+fluid_width, height+fluid_width)), pcd.remove_statistical_outliers(nb_neighbors, std_ratio), pcd.remove_radius_outliers(nb_points, radius)
• Compare: threshold, noise_threshold

For this specific project, use my modified version of these packages instead of the original ones in the catkin workspace: universal_robot, Universal_Robots_ROS_Driver, realsense-ros.

The position of printer_link defined in catkin_ws/src/universal_robot/ur_description/urdf/inc/ur_macro.xacro may not be accurate and requires manual adjustment. To do this, move the robot's end effector, which is set to be printer_link, to a position. Capture and visualize a frame in Open3D with the printer attached to the robot. In the window, also include a point indicating the goal position. Compare and adjust the translations in ur_macro.xacro accordingly, so that the printer's tip and the target point align. This step is necessary because the point cloud comparison relies grealy on the alignment of the printed object and the ideal model.

Due to the unsteadiness of camera holder, tape the camera firmly to the holder before starting the experiment, and recalibrate anytime you feel necessary.

If the IK solution of robot shown in RViz distorts greatly resulting in collision during movement, the most effetive way is to examine and modify joint degree constraints in catkin_ws/src/universal_robot/ur_description/config/ur10e/joint_limits.yaml.

References

• Open3D
• IntelRealSense
• easy_handeye
• Universal_Robots_ROS_Driver
• universal_robot