3 | Project of the Mechanical Engineering Competence Centre “Development of compact container cells of robot manipulators for processing of heterogeneous objects”

Activity 1: Industrial research

1. An analysis of the food industry has been carried out and the stage of the chain to be subjected to automation has been identified – in-depth research of sorting for recycling and re-use of used packaging, identifying sorting requirements and problems.   
2. Products, materials and their characteristics have been studied.   
3. Techniques and tools have been studied to identify the best possible mix of techniques and tools for efficient separation of products.   
4. The general architecture and principles of the machine learning flow have been developed.   
5. The requirements for the subsystem of operational control signals of robot manipulators have been defined.   
6. The requirements for associated robot manipulator operating environment context retrieval sensors have been defined.

Main conclusions:

• Recognition and sorting of very heterogeneous objects (food packaging), the composition of which is different in colour, shape, size, material, have been identified as a problem to be solved.   
• The solution should include computer vision and machine learning algorithms that take over a significant part of the decisions to be taken, because during a working shift a person gradually loses the ability to detect deviations from the standard or feature to determine compliance with specific category, to perform sorting.   
• The solution should include a collaborative flow that includes continuous replenishment (training) of the knowledge base used in the operation of the machine learning algorithm by receiving information about new objects, because new types of packaging are continuously appearing in the food industry.   
• The solution should include sorting methods, various sensors (spectrometer, weight cell) and mechatronic elements (conveyor belt, robot manipulator), alighting the requirements for accuracy and effectiveness of the sorting result with the costs of the elements concerned. For optimal results, methods and elements can be combined, the process can be partially automated as the end user performs specific object preparation activities.

Activity 2: Experimental development:

1. Work on the creation of the machine learning flow has started.

Work on identification of objects of the test dataset taking into account physical and visual properties of packaging materials.