Release Notes#
Click each tab to learn about the new and updated features in each release of Intel® Robot Vision & Control Framework.
RVC v2.1 release includes bug and security updates as well as Intel Lab’s Histodepth Pointcloud Segmentation algorithm in a Virtual Fence application.
Intel Lab’s Histodepth Pointcloud Segmentation Virtual Fence: Now part of the RVC package, this virtual fence application running on ROS uses depth information from an Intel RealSense camera to create dynamic and static scene segmentation maps to enable live robotic virtual fencing and safety bounding. The use of this segmentation algorithm enables a drop-in approach to virtual fencing, requiring no training or learning before deployment.
Features
New dynamic path planning algorithm available (Dobby path planner available under NDA - contact eci.maintainer@intel.com for details).
New dynamic virtual fencing algorithm available.
Known Limitations and Issues
RVC v2.0 provides a SW framework to control a robot based on vision based object detection. It’s a collection of SW components that can be deployed provided a debian packages. But it is expected to be customized, hardened, optimized and extended by customers before it’s used as a final product.
RVC focuses on demonstrating consolidation of the following functionalities:
Camera input
Vision based object classification and 3D pose detection.
Trajectory calculation to detected object pose.
Robot control
Features
Defined API between vision and control.
3D detection and 3D pose detection components.
2.5D detection and pose detection components.
Control of ROS2 supporting Robots.
Control of ROS2 not supported robots (example implementation for UR5).
Simple static and dynamic use case example.
Known Limitations and Issues
Debian packages:
Package |
Deb Package |
Description |
|---|---|---|
gui-settings |
|
Custom message for GUI to RVC communication Package. |
moveit2-servo-motion-controller |
|
RVC Motion controller interface-based plugin implemented using moveit2 servo. |
non-oriented-grasp |
|
Non oriented grasp plugin. |
openvino-inference-plugin |
|
RVC AI interface-based Plugin implementing yolo V5, v6 and V8 using openvino inferencing. |
oriented-grasp |
|
Non oriented grasp plugin. |
robotiq-controllers |
|
Provides controllers for robotiq 2f gripper. |
robotiq-driver-plugin |
|
Robotiq 2f driver. |
rvc-ai-interface |
|
Interface every 2.5D AI plugin shall inmplement. |
rvc-dynamic-motion-controller-use-case |
|
Main package for RVC motion controller. |
rvc-grasp-interface |
|
Pluginlib interface that a pose to grasp plugin has to implement. |
rvc-messages |
|
RVC vision interface messages. |
rvc-motion-controller-interface |
|
Interface defining how motion controller plugins shall be implemented. |
rvc-object-detection-engine |
|
AI inference pipeling Stage 1 object detection engine, plugin based to allow different AI models, and devices. |
rvc-packages-all |
|
Metapackage aggreating all RVC packages. |
rvc-panel-rviz2-plugin |
|
RVC controls for RViz. |
rvc-pose-detector |
|
AI pose alignment stage Package. |
rvc-profiler |
|
Node to gather various statistics of RVC. |
rvc-rotated-object-detection |
|
Rotated object detection using ORB features. |
rvc-static-motion-controller-use-case |
|
Main package for RVC motion controller. |
rvc-vision-main |
|
Start up and configuration files for the whole vision component. |
rvc-vision-messages |
|
Custom message between AI and motion controller. |
state-machine-msgs |
|
Custom message for state machines communication Package. |
trac-ik-kinematics-plugin |
|
A MoveIt! Kinematics plugin using TRAC-IK. |
trac-ik-lib |
|
TRAC-IK is a faster, significantly more reliable drop-in replacement for KDL’s pseudoinverse Jacobian solver. |
ur-pendant-motion-controller |
|
Direct Universal Robot pendant controller. |
Initial release of Robot Vision & Control Framework (RVC). RVC allows closed loop automatic object recognition and robot manipulation of a set of moving objects.
Features
Six Degree of Freedom (6DoF) real-time object detection
Manipulator dynamic tracking, that is, robot adjusts trajectory towards target in real-time
Known Limitations and Issues
The object set must be known prior, that is, the objects need to be present on file system in the pointcloud format, matching the real objects
Camera position must be accurate by either adjusting the camera in a particular position or changing the position in a configuration file
Only the Universal Robots™ family has been tested (more specifically, UR5e), the Robotiq 2F-85 Gripper and D415 Intel® RealSense™ Camera.
Only one object can be placed under the camera. Multiple objects could work, however there are known issues. For this release, make sure that there is only one object under the camera or none.