Modify Hardware Setup#

RVC Framework has been designed with modularity and abstraction in mind. In theory, this means that every hardware component could be substituted. However, due to the diverse range of devices and control mechanisms, the specific modifications required for seamless operation may vary considerably.

Change Camera#

This section explains the modifications required if you use different cameras:

Use a camera of same model#

To use a camera of the same model, such as Intel® RealSense™ D415, but a different location, make the following change in the <usecaseinstallationpath>/urdf/d415camera<namespace>.xacro file to reflect the real position:

<origin xyz="0.330 0.7210 0.755" rpy="-3.141592654 1.570796327 -1.570796327"/>

Note

xyz are in meters and roll pitch and yaw angles are in radians.

Note

To help the Vision component, measure the distance between the center of camera and the first flat surface (in our case the conveyor belt surface) and modify the value of the z_threshold (in meters) parameter in the <visioncomponentsinstallpath>/rvc_pose_detector/config/parameters.yaml. If z_threshold is closer to the camera than the maximum distance an object could be, the object will be cropped and never be detected.

Use Different Stereo Camera#

Note: Only Intel® RealSense™ cameras have been tested. To adapt the RVC Framework to a different camera, note that the ROS 2 node handling the new camera must expose a pointcloud ordered (that is, the pointcloud is not monodimensional array, but must be a 640x480 pointcloud and ordered). This is necessary to extract the cropped pointcloud from the bounding box coming from AI, which is done on a 2-D RGB channel.

To use a different stereo camera, edit the <usecaseinstallationpath>/launch/robot_demo_launch.py file and replace the following lines with the urdf and xacro from the gripper manufacturer:

camera_xacro_path = get_package_share_directory(<usecasepackagename>) + "/urdf/d415camera<namespace>.xacro"

This xacro generates a Universal Robot Descriptor File (URDF) that ROS2 uses to connect all elements together and perform necessary transformations from one reference system to another.

In particular, the poses come in the camera reference system, and the manipulator operates in world reference. So, the pose reference system and the details of how and where the camera is physically connected in the world is critical. The ROS2 node of the new camera will provide a URDF file, which must be connected to the world reference link.

This linking is done in d415camera1.xacro, which can be used as a reference. In the RVC demonstration, the base of the camera is at xyz="0.330 0.7210 0.755", rpy ="-PI,PI/2,-PI/2" to point to the right camera xacro that the manufacturer provides.

Additionally, make modifications on the AI side, such as removing the RealSense ros2 node and replacing with a ROS 2 of the new camera.

The AI publishes the detected pose with:

pose.header.frame_id = "camera_depth_optical_frame";

Modify this value according to the URDF of the new camera.

Use Different Non-stereo Camera#

To use a different non-stereo camera or an entirely different object detection technology, drop the RVC object recognition, keeping in mind that the RVC Robot Motion Controller is tracking a topic configured in src/robot_demo_main/config/parameters.yaml as object_pose_topic: "object_poses" of type rvc_messages::msg::PoseStamped as defined in src/shared_messages/rvc_messages/msg/PoseStamped.msg.

Change Universal Robot Model#

The default robot model is UR5e. If you are using a different robot model from the same family, do either one of the following:

  • Modify ur_type_arg = LaunchConfiguration in the launch file

    <dynamicusecaseinstallationpath>/launch/dynamic_demo_launch.py.

    The default value of ur_type is ur5e.

  • Append ur_type in the launch command line:

    ros2 launch rvc_dynamic_motion_controller_use_case dynamic_demo_launch.py robot_ip:=<ROBOT_IP> ur_type:=<URMODEL> &

Change Gripper and Manipulator#

The current implementation handles a composite robot, which is a manipulator connected to a gripper as it is a single robot with seven joints: six joints from the UR5e and gripper actuator as seventh joint.

To do so, RVC needs a composite URDF file, which includes the original UR5e URDF, the Robotiq URDF, and the joint connecting the UR5e end effector link (ee_link) to the gripper base link.

Moreover, the Robotiq original ROS1 URDF used in the demonstration did not include the ros2_control specification to expose the “joint” to the ROS2 hardware driver abstraction. Hence, it was added.

You can find the complete URDF glue file at <dynamicusecaseinstallationpath>/urdf/composite_u5_robotiq_2f_gripper.xacro. Use this glue file as a reference to compose a new glue file when replacing the robot and gripper.

Change the following:

  • The xacro ur.urdf.xacro include directive for the robot

  • The robotiq_arg2f_85_model_macro.xacro directive for the gripper.

If the ROS2 driver of the new gripper contains the ros2_control, do not add it here.

It is important to concatenate (link) the last link of the chain for the manipulator and the first link of the gripper and connect them in the new composition URDF. In this demonstration, for the ur5e model, the last link is tool0 and the first link of the gripper is robotiq_arg2f_base_link. In the composite_u5_robotiq_2f_gripper.xacro, a new link ee_link connecting the two was created.

Note: The center between the two fingers must be positioned on the target pose to pick the object up. The Robotiq gripper URDF does not specify any joint or link that can be used for computing the inverse kinematics of this point. So, the ee_link is offset by the z component at 0.174 meters from the parent, which is the end effector of the UR5e. The value 0.174 is exactly the length of the gripper base and the point where the fingertips are at when the gripper is closed.

The following figure shows the current global URDF representation as seen by ROS2 RQt tf_tree tool.

TFTree

Figure : TFTree#

In the graph, blue indicates the URDF coming from Robotiq gripper, green from Universal Robot, and red from RealSense ROS2 node.

This tool validates whether:

  • All links and joints are interconnected

  • The graph is not disconnected anywhere

  • The camera base connects to the world

  • The manipulator base connects to world

  • The gripper connects to the end effector of the manipulator.

If you physically connect the camera to any robot link instead of on a fixed location, the graph will represent it as the camera tree being connected to that link.

Change Manipulator without supported drivers#

RVC is based on ROS 2 framework, hence its able to drive manupulators with ROS 2 support. This means the manufacturer has to provide a ROS 2 driver, fully implementing the ROS 2 Control interfaces and MoveIt 2 support.

If the robot is not supported, there is another way, albeit potentially not straight forward: