Melodic_(TurtleBot3)Grasp Automatically

Tutorial using ROS for ver.3

Melodic Edition Backup

Grasping with TurtleBot3 automatically †

This example is a simple example that automatically grasps an object.

TurtleBot3 has OpenManipulator.

Object recognition by darknet_ros (YOLO) and Point Cloud data are used.
So CUDA must be installed in the Ubuntu in order to recognize objects using YOLO.
Although it is possible to check this example scene using CPU without installing CUDA, the object recognition speed is very slow.

The outline of the operation is as follows.

1. Move the TurtleBot3 to the front of the target object by keyboard operation.
2. Move the TurtleBot3 so that the target object comes to the center of the color image.
3. Enter a number to specify the object to be grasped.

When the above operation is performed, TurtleBot3 tries to grasp the object as follows:

1. The darknet_ros (YOLO) performs object recognition using a color image.
2. Estimate the 3D position of the target object using color images and Point Cloud information.
3. Grasp the calculated 3D position.

Please refer to here for the specifications of Open Manipulator.
Please refer to here for camera specifications.
Because it is convenient to shorten the minimum distance of the depth sensor, RealSense SR300 is used instead of RealSense R200.

Build Ubuntu environment †

In this example, darknet_ros (YOLO) needs to be installed in Ubuntu environment.
It is recommended to check the operation of normal darknet without using ros.

The procedure is as follows:

1. Download and install CUDA from here. (As mentioned above, it is also possible to run using the CPU without installing CUDA)
   (For details, please adjust according to your PC and NVIDIA Driver environment.)
2. Git clone darknet_ros. (Reference)
   Download the specific commit on the melodic branch.

   $ cd ~/catkin_ws/src
   $ git clone --recursive -b melodic https://github.com/leggedrobotics/darknet_ros.git
   $ cd darknet_ros
   $ git checkout b9d9a7dd

3. Install darknet_ros. (Reference)

   $ cd ~/catkin_ws
   $ catkin_make -DCMAKE_BUILD_TYPE=Release

   * Depending on the combination of GPU and CUDA, you may get an error like "Unsupported gpu architecture 'compute_30'" during catkin_make.
   In that case, you can comment out the line related to compute_30 in CMakeLists.txt of darknet_ros to avoid the error.

Startup Procedure †

Start the Ubuntu side and then the Windows side.

Ubuntu side startup procedure †

Open a new terminal and run the following command:

$ roslaunch sigverse_turtlebot3_open_manipulator grasping_auto.launch

Windows side startup procedure †

Start the [Assets/SIGVerse/ExampleScenes/Turtlebot3/OpenManipulatorSR300(.unity)] scene with reference to here.

Run †

You can control TurtleBot3 by keyboard operation on the grasping_auto terminal.

* Check the terminal for details of the operation.

* Because of the high image processing load, the robot's camera image transmission interval is set to 1000 [ms].
(If you want to shorten the sending interval, please reduce the Sending Interval of the TurtleBot3PubSR300RGBController attached to turtlebot3_with_open_manipulator_SR300/RosBridgeScripts)

If you want to finish, stop the Unity side and then the ROS side.

Ubuntu side (The details may differ from the latest version)

#ref(): File not found: "TurtleBot3GraspAutoUbuntu.png" at page "Melodic_(TurtleBot3)Grasp Automatically"

Windows side (The details may differ from the latest version)

#ref(): File not found: "TurtleBot3GraspAutoWindows.png" at page "Melodic_(TurtleBot3)Grasp Automatically"

Windows side (grasped "clock") (The details may differ from the latest version)

#ref(): File not found: "TurtleBot3GraspAutoWindowsGrasping.png" at page "Melodic_(TurtleBot3)Grasp Automatically"

Reference Video (Image recognition is slow because no GPU is used.)