Upravljanje SofIA robotskom hvataljkom temeljeno na dubokim neuronskim mrežama

undergraduate thesis

• Title (english): Deep learning based control of robotic gripper SofIA
• Author: Petar Sorić
• Mentor: Matko Orsag (mentor)
• Committee member: Matko Orsag (član povjerenstva)
• Committee member: Jelena Vuletić (član povjerenstva)
• Committee member: Robert Milijaš (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-07-07, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U ovom radu opisan je postupak razvoja jedne konvolucijske neuronske mreže koja uspješno klasificira slike pribavljene sa robotske hvataljke SofIA. Napravljen je pregled nekoliko radova iz područja "soft" robotike koji se bave sličnom tematikom. Opisani su koraci generiranja sintetskih podataka za treniranje u programu Blender i prikazana je struktura tih podataka. U daljnjim poglavljima opisana je struktura konvolucijske neuronske mreže i njena programska implementacija. Neuronska mreža trenirana je koristeći 3 različite kombinacije sintetskih i pravih podataka za treniranje, za svaku kombinaciju testirali smo neuronsku mrežu na pravim podacima. Odabrali smo najbolje rješenje koje ima približno 95% točnosti na pravim podacima. Usporedili smo ostala rješenja s najboljim i komentirali rezultate.

Abstract (english)

This paper describes the development of a convolutional neural network used for classification of pictures obtained from SofIA robotic gripper. A brief overview of the soft robotics research area is written, commenting on a few papers researching soft robotic grippers. The process of generating the synthethic dataset in Blender and it's structure is described in detail in the next chapter. After generating the dataset used for training the neural network, the structure and specific parameters of the network is described. The code and other tools used for designing, training and testing the neural network is shown next. There were 3 different training methods used, a combination of synthethic and real data. The best of the three approaches was picked as the final solution, which had a 95% accuracy when tested on a real sample.Finally, the three approaches were compared based on their accuracy and the results were discussed.

• Keywords: duboko učenje; konvolucijska neuronska mreža; robotska hvataljka
• Keywords (english): deep learning; convolutional neural network; robotic gripper
• Language: croatian
• URN:NBN: urn:nbn:hr:168:154177
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-15 11:14:28