The rc_reason ItemPickAI software enables the efficient separation and oriented placement of so-called ‘unseen’ objects – i.e. objects that the system has not explicitly learned by teaching models or seen in training data beforehand.
This AI-based solution for robotic pick-and-place applications with suction grippers calculates aligned grasp poses for a suction gripper on unknown, deformable objects of a certain category. These objects can come in chaotic boxes or load carriers, and both mixed and unmixed configurations are possible.
During commissioning, only the object category (e.g. ‘bag’) and the size of the gripper’s suction cup need to be selected. These parameters can be flexibly adapted to new conditions at any time. This means that pick-and-place applications such as picking and (de)palletizing can be implemented efficiently within a very short time frame, even without expert knowledge in the field of AI or image processing.
The rc_reason ItemPickAI module can be used offboard with any 3D sensor via Roboception’s rc_cube and can be configured and read out via the standard interface. ItemPickAI calculates a configurable number of grasp poses for a suction device on all identified objects within a predefined workspace.
Robust segmentation of objects with uneven surfaces and varying, deformable content is made possible by the use of neural networks. This ensures that a grasp point is always placed in the center of the segmented objects. In addition, the orientation of the object is determined to enable an oriented placement.
As the size of the selected suction device can be defined individually, the ItemPickAI software can be used with any standard suction gripper.
The working area is defined either automatically (e.g. by recognizing a container) or manually (by defining the area of interest). Sample programs facilitate integration with robot controllers.
For ItemPickAI applications, the use of the rc_randomdot projector is recommended, especially when load carriers need to be detected within the pick process.
| Hardware requirements | Compatible 3D stereo sensor, requires use of rc_cube |
| Grasp computation time | Grasp point computation time <1 second (depending on object and scene) |
For more detailed information about our products and our software modules for your robot vision solution, please visit our Knowledge Base:
You would like to find out whether our portfolio is suitable for your robotic application? Simply request a feasibility study free of charge, and get a live demo of our products.
Would you like to try out one of our sensors and software solutions? Our Try-&-Buy-option gives you the chance to test our products before you decide, and to be sure you make the right choice for your application.
