As part of our thesis, we explored the possibility of using artificial intelligence language models, such as ChatGPT, to detect the location and orientation of objects in images, which we later used as a basis for obtaining trajectories for performing product transfer operations to the disposal site. First, we reviewed the theoretical foundations of artificial intelligence, machine vision, and API (Application Programming Interface) operation, and described Dobop Magician, which we used to perform product transfer operations. We then conducted experimental tests to verify whether artificial intelligence, in our case the ChatGPT model, can successfully locate objects in images and whether it is accurate enough to use this data for further use, or whether the robot can even hit the desired target. To do this, we first had to connect the robot to the program via API. Finally, we tested the accuracy of the system and found that it is satisfactorily accurate, always pick and place the products correctly, but it needs to be calibrated well beforehand.
|
