Ciklum has been working on an algorithm which provides gripping point position and limb orientation of the pig. For effective gripping of exterior masses, particularly in the case of slaughter, it is important that the MFC has the capability to recognise a carcass and classify it from an anatomical perspective (e.g., head, limbs, back, belly, etc.). Furthermore, it is important to be able to continuously track the orientation and location of key parts, as warm meat is flexible, and will change shape or relocate while being manipulated. To address this, a software system for external classification has been developed.
Anton Popov from Ciklum explains: The biggest challenge was to work with the limited amount of experimental data due to the pandemic restrictions. That required additional actions in training deep neural networks to reach the appropriate performance in gripping points localization.
Future work consists of improving the quality of the algorithms and integrating the solution within the Meat Factory Cell.