Abstract:

In daily life, many auxiliary functions for humans have begun to be performed by robots. Robot arms, especially used in industry, identify the objects coming into the work area, determine the properties of the relevant object, and can make decisions close to human intelligence in production by adjusting the holding power and position that should be applied accordingly. Among these functions, recognition, positioning, and retention are performed by artificial intelligence algorithms based on people's thinking and solution generation systems. Deep learning, which is one of the artificial intelligence approaches, has been used for many different problems recently and it has been seen that successful results have been obtained. The aim of this study is to position the object in order to reach the holding position for a classical robot arm and to provide a system that can automatically determine the appropriate grip parameters for the object to be held and lifted. In the creation of the mechanical system of the robot arm, the models produced from 3D drawing programs were printed using a 3D printer. In this way, information on how to use 3D printer technology in such a system was shared. In addition, thanks to this technology, low cost, and easily configurable models have been created. A deep learning algorithm based on Convolutional Neural Networks (CNN), one of the artificial intelligence methods, has been used in order to realize the real-time object recognition function and to make the system intelligent and learnable. Detailed information is given about the preliminary preparation processes that must be done in order for the algorithm to work efficiently. From the results obtained with the simulation tests, it has been seen that the proposed system successfully performs these functions.

Keywords: