PoseStream Model

First, the training service involves programming in PoseStream Designer, where users configure the object models, gripper settings, and relevant parameters. This setup ensures the system is prepared to generate the necessary image data for the training process.

Step 2 of the training service requires uploading the files to our cloud service and clicking the start button. The system then generates a large volume of images, and within 24 hours, the customer can access the trained PoseStream model as the outcome.

Step 3 of the training service involves training the network to learn from the generated images, refining its ability to detect and estimate object poses. The final output of this process is a fully trained PoseStream model ready for deployment in robotic applications.

Step 4 of the training service involves downloading the trained PoseStream model to your Jetson device and viewing it in the PoseStream Viewer. To use PoseStream with your robot, a hand-eye calibration with a calibration board is required, followed by TCP (Tool Center Point) calibration for precise robot operation.

Step 5 of the training service focuses on Robot Communication. The robot sends a command to PoseStream to capture an image, and PoseStream returns the detected object poses and gripping positions for objects that can be picked without collision. Robot communication is built into PoseStream and works seamlessly with ABB and UR robots. For other robot brands, Cobotic offers custom communication integration upon request.